Dialogue at the Edge of Fatigue: Personalized Voice Assistant Strategies in Intelligent Driving Systems

Abstract

With the rapid development of intelligent transportation systems, voice assistants are increasingly integrated into driving environments, providing an effective means to mitigate the risks of fatigued driving. This study explored drivers’ interaction preferences with voice assistants under different fatigue states and proposed a fatigue-state-based dialogue-awakening mechanism. Using Grounded Theory and the Stimulus–Organism–Response (SOR) framework, in-depth interviews were conducted with 25 drivers from diverse occupational backgrounds. To validate the qualitative findings, a driving simulation experiment was carried out to examine the effects of different voice interaction styles on driver fatigue arousal across various fatigue levels. Results indicated that heavily fatigued drivers preferred highly stimulating and interactive voice communication; mildly fatigued drivers tended toward gentle and socially supportive dialogue; while drivers in a non-fatigued state preferred minimal voice interference, activating voice assistance only when necessary. Significant occupational differences were also observed: long-haul truck drivers emphasized practicality and safety in voice assistants, taxi drivers favored voice interactions combining navigation and social content, and private car owners preferred personalized and emotional support. This study enriches the theoretical understanding of fatigue-sensitive voice interactions and provides practical guidance for the adaptive design of intelligent voice assistants, promoting their application in driving safety.

1. Introduction

Fatigue driving has become one of the major threats to road traffic safety worldwide. Drivers in a fatigued state experience a range of physiological and cognitive impairments, such as decreased attention, slower reaction times, impaired judgment, and operational errors. These factors significantly increase the likelihood of traffic accidents [1,2]. Data indicate that fatigue driving accounts for 35% to 45% of road traffic accidents, resulting in approximately 1550 fatalities, 71,000 injuries, and an economic loss of USD 12.5 billion annually [3]. Professional driver groups—particularly long-haul truck drivers, taxi drivers, and individuals with long commuting hours—are especially susceptible to chronic fatigue due to irregular work schedules, limited opportunities for rest, and prolonged periods of high-intensity driving. These factors place them at high risk of fatigue-related driving incidents, and industry-specific attitudes and contextual factors further influence the prevalence of such behavior [4]. Consequently, accurately detecting driver fatigue and implementing effective intervention strategies has become a critical topic in interdisciplinary research spanning intelligent transportation systems, human factors’ engineering, and behavioral sciences.

At present, interventions for fatigue driving primarily focus on two aspects: fatigue state detection and real-time alert mechanisms. Fatigue detection technologies have evolved from early-stage subjective questionnaires and physiological indicator methods [5] to intelligent recognition systems centered on behavioral data and image processing. Features such as electroencephalogram (EEG) signals [6], eye-tracking [7], facial expression analysis [8], and driving behavior [9] have been widely employed in the development of fatigue detection models. The integration of deep learning algorithms has significantly enhanced the accuracy and timeliness of these systems. For instance, a recent study proposed a novel method combining deep convolutional neural networks with emotional state analysis to detect driver fatigue, improving the safety and reliability of intelligent driving assistance systems under varying lighting conditions [10]. In addition, several automotive companies have introduced integrated driving assistance systems. Cooperative Advanced Driver Assistance Systems (CADAS), for example, enhance road safety and mitigate fatigue by providing traffic information several kilometers ahead [11]. However, most existing systems rely on fixed-threshold judgments and uniform alert strategies, lacking the capacity to perceive drivers’ psychological states and interaction preferences. This often results in issues such as “over-reminding” or “delayed response”, ultimately affecting user acceptance.

In addition to detection technologies, multimodal stimulation-based intervention approaches have become a key focus of research. Scholars have explored non-contact sensory stimuli—such as auditory, olfactory, and visual cues—to awaken drivers. In terms of auditory stimulation, music has been shown to enhance task performance during driving and effectively increase driver alertness [12,13]. For olfactory stimulation, studies have found that refreshing scents such as peppermint and citrus can reduce subjective feelings of fatigue and delay the onset of fatigue [14,15,16]. Visual interventions, including speed alerts and flashing warnings in the mid-peripheral visual field, have also been confirmed to improve wakefulness during driving [17,18]. The i-DREAMS fatigue intervention strategy aims to keep drivers within a safe driving zone by providing both real-time and post-trip interventions, incorporating gamification elements to enhance engagement [19]. However, these stimulation-based interventions have three main limitations: (1) short-lasting effects that lead to adaptation-related fatigue; (2) difficulty in controlling stimulation intensity, with significant individual variability in responses; and (3) potential for frequent stimuli to provoke annoyance or emotional fluctuations, thereby interfering with normal driving behavior. As a result, there is an urgent need for a more flexible, dynamic, and personalized intervention approach to enhance effectiveness.

In recent years, voice interaction has emerged as a natural and convenient means of communication, widely adopted in intelligent driving scenarios. Studies have shown that, compared to single-turn dialogues, users in driving contexts prefer multi-turn interactions, pointing toward the need for more interactive in-vehicle voice assistants [20]. Moreover, due to its eyes-free and hands-free nature, voice interaction holds unique potential for fatigue intervention by providing cognitive stimulation to enhance driver alertness [21,22]. While existing research has made progress in developing voice assistant functions and exploring their role in emotional regulation [23,24], significant gaps remain. On the one hand, most systems still rely on templated, one-way interactions and lack dynamic perception of the driver’s fatigue level, cognitive load, and interaction preferences. On the other hand, current studies primarily focus on system performance or emotional expression itself, with limited exploration of users’ subjective experiences and behavioral response mechanisms.

To address these gaps, this study focuses on fatigue-driving intervention scenarios and proposes a dialogue-awakening mechanism centered on the driver’s fatigue state. Unlike conventional system-oriented approaches, this research combines Grounded Theory with the Stimulus–Organism–Response (SOR) framework to deeply analyze how drivers’ preferences for dialogue content, tone, frequency, and style vary under different levels of fatigue. Through in-depth interviews and coding analysis of representative driver groups, we identify heterogeneous demands for voice assistant functions. Based on these findings, we design a driving simulation experiment to validate the interview results, thereby providing both theoretical grounding and empirical evidence for the development of context-aware and personalized in-vehicle voice interaction systems.

2. Methods

2.1. Experimental Design

This study adopts Grounded Theory as a qualitative research approach, integrated with the SOR theoretical framework, to explore drivers’ interaction preferences with voice assistants and the underlying dialogue-awakening mechanisms under varying fatigue states. As an inductive methodology, Grounded Theory emphasizes theory generation from empirical data rather than relying on preconceived assumptions, thereby enhancing the scientific validity and practical relevance of the findings. Through systematic analysis of interview data, this approach facilitates the construction of a theoretical model that captures the dynamic and complex nature of drivers’ interaction behaviors with voice assistants.

Within the framework of Grounded Theory, this study follows a three-stage coding process: open coding, axial coding, and selective coding. Initially, open coding is used to conduct a preliminary analysis of the interview transcripts, identifying key words and behavioral patterns. Axial coding then organizes these elements into categories reflecting drivers’ interaction needs and preferences across different fatigue states. Finally, selective coding integrates these categories into a coherent theoretical model [25].

To further understand the mechanisms of driver responses under different fatigue conditions, this study incorporates the SOR framework [26]. In this model, Stimulus refers to external factors influencing driver decision-making, such as the road environment, fatigue level, and the content and frequency of voice assistant interactions. Organism represents the internal psychological and physiological states of the driver—fatigue severity, cognitive load, and alertness—that mediate information processing. Response captures the driver’s behavioral reactions to the voice assistant, including acceptance, interaction needs, and preferences regarding content, tone, and speech rate.

2.2. Interview Participants

The interview participants in this study were recruited from a driver database maintained by key transportation enterprises and ride-hailing platforms in Fujian Province, China. Additional samples of private car owners were added as supplements through acquaintance referrals and on-site recruitment. Preliminary screening criteria included holding a valid driver’s license, having at least two years of driving experience, and bearing no record of major traffic accidents in the recent past. Based on the research objectives, three typical categories of drivers were prioritized: long-haul freight drivers, taxi/ride-hailing drivers, and private car owners. These groups differ significantly in terms of driving task characteristics, fatigue triggers, and expectations for voice assistant functionality, providing a multidimensional perspective on personalized voice interaction preferences under fatigue-driving scenarios.

The selection of interviewees followed the procedures of theoretical sampling. In the initial stage, the research team randomly selected drivers from the defined groups to conduct in-depth interviews and performed open coding to identify preliminary concepts. As data analysis progressed, axial and selective coding were used to explore relationships among categories and identify theoretical gaps. Accordingly, the sample structure was adjusted, and participants with specific attributes were purposefully recruited to enrich theoretical development and achieve conceptual saturation [27].

A total of 25 drivers were interviewed for this study, including 9 long-haul freight drivers, 8 taxi/ride-hailing drivers, and 8 private car owners (

Table 1. Distribution of sample characteristics of interviewed drivers.

ID	Occupation Type	Gender	Age	Driving Experience (Years)	Primary Driving Scenarios	Driving Time Characteristics
P01	Long-haul truck driver	Male	45	15	Expressways, logistics routes	Frequent night driving, long durations
P02	Long-haul truck driver	Male	35	10	Expressways, national highways	Long continuous driving time
P03	Long-haul truck driver	Male	46	18	Inter-provincial trunk transport	High fatigue frequency
P04	Long-haul truck driver	Male	58	25	Expressways	Frequent night driving
P05	Long-haul truck driver	Male	38	9	Expressways, county-level logistics	Irregular driving hours
P06	Long-haul truck driver	Male	49	20	Expressways, mountain roads	High driving intensity
P07	Long-haul truck driver	Male	59	30	Long-distance national highways	Single driving sessions over 8 h
P08	Long-haul truck driver	Male	44	16	Expressways, urban outskirts	Frequently fatigued during night driving
P09	Long-haul truck driver	Male	36	12	Expressways, container transport	Frequent night operations
P10	Taxi driver	Male	48	20	Urban roads	Drives during both peak daytime and night hours
P11	Taxi driver	Male	37	12	Urban and suburban areas	Frequent passenger interaction
P12	Taxi driver	Male	55	28	Main urban roads	Moderate driving intensity
P13	Ride-hailing driver	Male	34	8	Urban expressways, residential areas	Flexible driving hours
P14	Ride-hailing driver	Female	33	7	Urban roads	Prefers daytime operations
P15	Ride-hailing driver	Male	45	15	Urban roads and nearby towns	Prone to nighttime fatigue
P16	Taxi driver	Male	58	22	Urban roads	Long working hours
P17	Ride-hailing driver	Female	38	10	City and suburbs	Prefers chatting with passengers
P18	Private car owner	Male	26	5	Urban commuting, short-distance travel	Occasional night driving
P19	Private car owner	Female	35	8	Commuting	Mostly daytime driving
P20	Private car owner	Female	46	15	Urban and nearby cities and counties	Mainly for family use
P21	Private car owner	Male	28	3	Urban roads	Prefers listening to music while driving
P22	Private car owner	Female	36	10	Commuting and family outings	Occasional self-driving trips
P23	Private car owner	Male	48	18	Expressways, national highways, road trips	Extensive long-distance driving experience
P24	Private car owner	Female	27	4	Urban and short-distance travel	Frequently uses voice assistant
P25	Private car owner	Male	34	6	Urban driving and holiday travel	Strong preference for emotion-regulating voice interaction

). Efforts were made to ensure diversity and comparability in gender, age, years of driving experience, and driving scenarios. Throughout the interview process, strict ethical standards were followed. All participants signed informed consent forms and were clearly informed of the research purpose and their rights. This sampling approach helped ensure the breadth and depth of the data, enhancing the overall rigor and validity of the findings [28].

2.3. Data Collection

To gain deeper insight into the dialogue-awakening mechanism under fatigue driving conditions, this study conducted one-on-one, in-depth semi-structured interviews with 25 drivers. The participants were drawn from three representative driver groups—long-haul truck drivers, taxi and ride-hailing drivers, and private car owners—to ensure diversity and representativeness in the data. Informed consent was obtained from all participants prior to the interviews, and all procedures adhered to ethical research standards. The interviews focused on themes including dialogue habits, interaction needs, and awakening effects of voice interaction. A detailed overview of the interview topics is presented in

Table 2. Interview outline on voice assistant interaction and fatigue driving.

Interview Topics	Outline Content
Conversation Habits	Do you use in-car voice assistants? In which situations do you use them the most?
	Are you inclined to initiate conversations with the voice assistant?
	Do you have any preferences regarding the voice assistant’s tone and pitch?
	Do you think the voice assistant is helpful for driving? In what ways?
Conversation Needs	When you are fatigued while driving, what kind of help would you like the voice assistant to provide?
	Do you need more reminders, companionship, entertainment, or navigation functions?
	What are your expectations for the voice assistant in terms of alleviating fatigue?
	Would you like the voice assistant to be able to detect your fatigue state?
Conversation Wake-up Effect	Have you ever been awakened by the voice assistant when driving fatigued? What was your reaction at the time?
	What type of voice content do you think is most effective for waking you up?
	Between gentle reminders, humorous banter, and strong commands, which do you prefer?
	Have you ever ignored the voice assistant? Why?

.

To ensure data quality, all interviews were audio recorded, with durations controlled between 30 and 50 min. After each session, the recordings were transcribed verbatim by the research team. During transcription, preliminary annotation and de-identification of sensitive information were conducted simultaneously to maintain participant confidentiality. In the subsequent open-coding phase of Grounded Theory, the transcribed texts were subjected to content analysis to extract concepts and initial categories, laying the foundation for the stepwise development of the theoretical framework.

2.4. Data Analysis

This study adopted a combined analytical strategy integrating Grounded Theory and the SOR framework to systematically explore the dialogue-based arousal mechanism under fatigued driving conditions. The data analysis proceeded in two main stages: (1) a coding process guided by Grounded Theory, and (2) an inductive analytical framework informed by the SOR theory.

2.4.1. Coding Process

To ensure systematic and in-depth data analysis, this study followed the three-stage coding procedure of Grounded Theory to process the interview data [29]:

Open coding: The researchers first conducted a line-by-line and paragraph-by-paragraph review of the 25 semi-structured interview transcripts, identifying and labeling expressions related to behaviors, emotions, perceptions, and experiences. The objective of this stage was to capture all potential concepts and patterns associated with fatigued driving, thereby forming a preliminary conceptual framework for further analysis.

Axial coding: Building on the open coding, the extracted concepts were categorized and synthesized to identify internal relationships and to construct more refined categories. This stage focused on identifying key factors related to voice interaction in the context of fatigue driving, such as dialogue types, voice characteristics, driver needs, and behavioral responses, as well as analyzing how these elements collectively influence driver alertness and fatigue states.

Selective coding: Based on the results of the previous two stages, the analysis concentrated on developing a core category centered on “the arousal function of voice interaction in fatigued driving”. In this stage, all coding results were integrated into a comprehensive conceptual model, from which the core elements of the dialogue-based arousal mechanism were distilled to form an explanatory theoretical framework.

2.4.2. The SOR Theory Framework

After completing the Grounded Theory coding process, this study employed the SOR model for inductive data analysis and framework development. The SOR theory provides a behavioral psychology perspective, categorizing the interaction between the driver and voice dialogue into three dimensions [30]:

Stimulus (S): In this study, voice dialogue is considered an external stimulus, encompassing the voice characteristics of the voice assistant, the type of dialogue content, and the interaction methods (e.g., humorous Q&A, casual conversation, task-oriented dialogue). These dialogue features, as external stimuli, directly influence the driver’s emotional and cognitive states.

Organism (O): This represents the internal state of the driver, including fatigue levels, attention levels, emotional states, and individual psychological needs. In this study, the organism level involves the driver’s physiological fatigue, psychological fatigue, and personalized dialogue requirements. Different driver groups (e.g., long-distance truck drivers, taxi drivers, private car owners) have varying levels of fatigue and dialogue needs, which influence their responses to voice interactions.

Response (R): The response refers to the behavior or psychological reaction exhibited by the driver after receiving the voice dialogue stimulus, primarily reflected in changes in driver alertness (e.g., attention restoration, increased fatigue). Based on the driver’s responses, this study categorizes the dialogue arousal mechanism into effective and ineffective responses, further analyzing the impact of different types of dialogue on driver alertness and safety.

3. Interview Results

3.1. Encoding Results

In this study, we adopted Grounded Theory to conduct step-by-step coding of the interview data, divided into three stages: open coding, axial coding, and selective coding. During the open-coding stage, we identified 38 initial concepts from the interview data, covering aspects such as the causes and impacts of fatigue driving, as well as drivers’ dialogue needs and preferences. In the axial-coding stage, these initial concepts were grouped into 13 main categories, further clarifying the internal relationships among them. Finally, in the selective-coding stage, through analyzing the main categories, we extracted six core themes that comprehensively reflect drivers’ dialogue demands and the adaptability of voice assistants during fatigue driving. These core themes include triggers of fatigue driving, dialogue needs, dialogue preferences, the impact of dialogue types, the adaptability of voice assistants, and directions for optimization, forming the basis for subsequent theory construction and strategy development (see

Table 3. Coding framework for voice interaction in fatigue driving based on Grounded Theory.

Selective Coding (Core Category)	Axial Coding (Subcategory)	Open Coding (Initial Concepts)
Fatigue Driving Triggers	Causes of Fatigue Driving	Long hours of driving, night driving, monotonous environment, driving pressure, physical fatigue, mental fatigue
	Effects of Fatigue	Cognitive fatigue, reduced attention, emotional fluctuations, degraded driving performance
Drivers’ Dialogue Needs	Dialogue Motivation	Active (refreshing, relieving boredom), passive (passenger interaction, work-related needs)
	Dialogue Targets	Familiar persons (family, friends), strangers (passengers), AI assistant
Drivers’ Dialogue Preferences	Preferences of Long-Haul Truck Drivers	Task-oriented (navigation, traffic info), highly interactive (Q&A)
	Preferences of Taxi Drivers	Social conversations (chatting with passengers)
	Preferences of Private Car Owners	Relaxed entertainment (music, news), family communication
Impact of Dialogue Types	Effective Dialogue Types	Humorous, highly interactive (e.g., Q&A, light chats)
	Ineffective or Fatigue-Inducing Types	Mechanical, repetitive, serious topics, emotionally neutral dialogues
Adaptability of Voice Assistants	Evaluation of Current Assistants	Rigid interaction, lack of personalization, excessive interruption
	Ideal Assistant Characteristics	Friendly tone, moderate speech rate, contextual adaptation, personalized recommendations
Optimization of Voice Assistants	Needs by Fatigue Level	Severe fatigue (strong stimuli), mild fatigue (light social interaction), no fatigue (minimal intervention)
	Intervention Frequency	Over-intervention, moderate reminders, demand-based adjustment

). In addition, we monitored theoretical saturation throughout the coding process. From the 22nd participant onward, the frequency of new concepts declined significantly, and by the 25th participant, no new concepts or categories emerged. Existing themes, such as tone and frequency in the “dialogue preferences” category, were merely repeated without extending the attributes of existing categories, indicating that theoretical saturation had been reached.

3.2. Fatigue Driving Triggers

In this study, the analysis of fatigue-driving triggers revealed a range of contributing factors that collectively influence drivers’ cognitive and physical states, ultimately leading to fatigue driving. Interviews with participants across different occupational groups and driving scenarios indicated that a prolonged driving duration, nighttime driving, monotonous environments, and psychological fatigue are among the primary factors triggering fatigue driving.

Prolonged driving was widely recognized as a major contributor to driver fatigue. Respondents consistently reported that during extended periods of driving—especially continuous driving for several hours—they tended to feel exhausted, with a diminished ability to maintain concentration. As participant P01 noted, “Driving for long hours at night makes my eyes feel blurry. I keep yawning and find it difficult to stay focused”. This phenomenon was particularly prevalent among long-haul truck drivers, especially during interprovincial trips or prolonged highway driving, where mental and physical endurance are frequently challenged.

Nighttime driving was identified as especially fatiguing. Participants reported that driving at night disrupts their circadian rhythms, requiring them to remain alert during natural fatigue peaks, which increases drowsiness. As participant P04 shared, “I often drive on the highway at night, and the drowsiness is very strong, especially in the early morning hours. It’s really hard to stay awake”. Night driving not only intensifies physical fatigue but also makes drivers more vulnerable to environmental distractions, negatively affecting reaction time and judgment.

Monotonous driving environments were also cited as a key trigger for fatigue. Respondents described how long periods of driving in environments with little variation—such as highways or mountain roads—lead to mental fatigue and reduced attentiveness. Participant P06 explained, “When I drive on the highway, especially in the mountains, the road conditions are very repetitive. After a while, I just can’t focus. It makes me really tired”. Such monotonous environments exacerbate driver fatigue and serve as an additional trigger for fatigue-related risks.

Psychological fatigue also played a significant role in shaping drivers’ fatigue states. Many drivers reported that in addition to physical exhaustion, they experienced mental strain due to heavy workloads and long driving shifts. As participant P15 stated, “The pressure from work is high. Sometimes, when I’m driving, my mind feels blank and mentally drained”. This psychological fatigue affects drivers’ emotional states and cognitive performance, further increasing the likelihood of fatigue driving.

3.3. Personalized Dialogue Needs of Drivers

Drivers’ personalized dialogue needs were analyzed in relation to their occupational type, driving scenarios, and fatigue status. Interview data revealed significant differences in dialogue preferences among different categories of drivers.

Long-haul truck drivers typically exhibited task-oriented dialogue needs, with a preference for highly interactive communication. For instance, P02 (a long-haul truck driver) stated, “During long-distance driving, what I need most is navigation and traffic updates, which help me stay focused”. Similarly, P06 remarked, “When I’m driving for extended periods, having interactive Q&A—especially about the weather or road conditions—helps me relax a bit”. These drivers rely on efficient information transmission during driving, and therefore prefer dialogues from voice assistants that are strongly task-focused, such as navigation and traffic information. Meanwhile, interactive elements help mitigate feelings of fatigue.

Taxi and ride-hailing drivers showed a stronger preference for socially oriented dialogues. Their interactions with passengers often provided a source of relaxation and social engagement. For example, P10 (a taxi driver) commented, “Chatting with passengers makes driving feel less monotonous, especially during night shifts. The interaction helps me relax mentally”. P17 (a ride-hailing driver) added, “Talking with passengers isn’t just about customer service—it’s social. That interaction helps reduce fatigue”. For these drivers, especially during peak hours or nighttime driving, social dialogue serves to alleviate psychological fatigue and reduce the sense of isolation.

Private car owners, in contrast, generally preferred more relaxed and entertainment-oriented dialogues. During commutes or short-distance trips, they tended to use voice assistants for casual conversations, music, or family-related communication. As P18 (a private car owner) noted, “I usually listen to music or news while driving—it helps me relax, especially during the morning rush. Music keeps me alert”. Likewise, P21 shared, “I enjoy chatting with my family while driving, especially on long trips. The voice assistant helps me stay connected with them”. Thus, private car owners’ dialogue preferences tend to focus on entertainment, relaxation, and family communication, enabling emotional regulation and social connection.

Drivers’ personalized dialogue needs vary based on their professional background, driving conditions, and fatigue levels. Long-haul truck drivers benefit most from task-oriented and highly interactive dialogues to maintain alertness and reduce fatigue. Taxi and ride-hailing drivers seek social interaction to combat loneliness and psychological fatigue, while private car owners prioritize relaxed, entertainment-based, and family-centered dialogues to enhance comfort and emotional well-being during driving. Voice assistant systems should therefore be tailored to accommodate these differentiated needs to optimize the user experience and promote driving safety.

3.4. The Impact of Dialogue Types on Fatigue Driving Awakening

In the context of fatigue driving, the type of dialogue between the driver and the voice assistant significantly affects the mitigation or exacerbation of the driver’s fatigue. Analysis of the interview data revealed that different types of dialogue have varying impacts on drivers’ fatigue states, which can be categorized into effective and ineffective dialogue types.

Effective dialogue types are typically characterized by strong interactivity, emotional relaxation, and cognitive stimulation. For example, P03 (a long-haul truck driver) stated, “When I’m feeling fatigued, if the voice assistant can make me laugh through humorous and lighthearted dialogue, it really wakes me up and driving doesn’t feel as exhausting”. Similarly, P09 (another long-haul truck driver) shared, “I enjoy interactive Q&A. When the voice assistant asks light, fun questions or shares interesting trivia, it breaks the monotony”. These highly interactive dialogues offer engaging content that captures the driver’s attention, effectively reducing cognitive fatigue and helping to maintain focus. This effect is especially pronounced during night driving or prolonged driving periods, where humor and Q&A-style conversations help regulate the driver’s mood and strengthen emotional connection during driving, thereby alleviating fatigue.

In contrast, ineffective dialogue types may intensify the driver’s sense of fatigue, particularly when the interaction is mechanical, repetitive, and emotionally neutral. P06 (a long-haul truck driver) noted, “If the voice assistant keeps repeating unnecessary messages or mechanically reminds me to drive safely, it feels like I’m being forced to listen, which actually makes me feel more tired”. P11 (a taxi driver) echoed this sentiment: “Sometimes the voice assistant says things in such a dull or serious tone, it feels like a lifeless machine, and that makes me even sleepier”. These types of mechanical and repetitive dialogues fail to provide emotional relief and may lead to decreased attention and heightened fatigue. Especially during extended driving sessions, the lack of emotional connection and interactivity can cause mental exhaustion and distraction, increasing the risk of fatigue-related driving incidents.

Overall, effective dialogue types help mitigate fatigue by offering emotional regulation and cognitive activation. Humorous and interactive conversations can positively influence drivers’ emotional states and attentional focus. Conversely, ineffective dialogue types, particularly those lacking emotional engagement, tend to worsen fatigue symptoms, leading to distraction and emotional disengagement. Therefore, in the design of voice assistant systems, it is crucial to incorporate emotionally rich and interactive dialogue content tailored to drivers’ fatigue levels to enhance their well-being and driving safety.

3.5. Adaptability and Optimization Requirements of Intelligent Voice Assistants

Drivers’ adaptability to intelligent voice assistants and their expectations for optimization are key factors in enhancing the effectiveness of fatigue driving prevention. Analysis of interview data revealed that the adaptability of voice assistants significantly influences drivers’ evaluations and directly impacts their fatigue-related behaviors.

Firstly, most drivers expressed dissatisfaction with current voice assistants, particularly regarding their interactive experience. Many reported that the assistants’ voice interactions felt rigid, lacked personalization, and sometimes became irritating due to excessive interruptions. For instance, P04 (a long-haul truck driver) remarked, “Sometimes the assistant sounds too robotic and asks very basic questions that don’t meet my actual needs—it just feels like a machine talking”. P16 (a taxi driver) echoed this concern: “The assistant keeps repeating the same information, which gets annoying. Especially during long drives, this repetition makes my fatigue feel even worse”. These responses indicate that current voice assistants lack flexibility and personalization, failing to adequately consider drivers’ real-time needs and emotional fluctuations.

Ideally, drivers expect voice assistants to have a friendlier tone, appropriate speech speed, context-aware responsiveness, and personalized recommendation functions. For example, P12 (a taxi driver) noted, “It would be great if the assistant could adjust its tone and pace based on how tired I sound—like using a softer voice when I’m fatigued”. P05 (a long-haul truck driver) hoped that the assistant could provide personalized suggestions based on road conditions, weather, and time, while minimizing unnecessary reminders and information overload. Such improvements would not only help reduce driver fatigue but also enhance the emotional connection between the driver and the assistant, thereby supporting better focus during driving.

Regarding optimization directions, different fatigue states entail different expectations for voice assistant behavior. For severely fatigued drivers, high-stimulation dialogue—such as humorous or lively interactions—is needed to energize them and reduce drowsiness. As P02 (a long-haul truck driver) explained, “When I’m really tired, if the assistant tells a joke or asks something funny, it helps me feel a bit less exhausted”. For drivers experiencing mild fatigue, low-intensity social dialogue—such as casual chatting or simple Q&A—can effectively relieve stress. P17 (a ride-hailing driver) commented, “I like having light interactions with the assistant, like chatting with a friend—it helps me relax a bit”. Meanwhile, for drivers without any obvious fatigue, the assistant should maintain minimal intervention to avoid unnecessary disturbance. P21 (a private car owner) remarked, “If I’m not tired, the assistant shouldn’t bother me too often—just give information when I need it”.

Additionally, the frequency of assistant interventions should be tailored to the driver’s fatigue level. For those who are severely fatigued, increased intervention may be necessary but should not excessively disrupt attention. For mildly fatigued drivers, moderate reminders can help alleviate fatigue. For those not experiencing fatigue, the assistant should maintain a low frequency of interaction, offering reminders only when needed. This demand-driven intelligent intervention strategy not only avoids unnecessary interference but also provides the most appropriate support according to the driver’s real-time state.

3.6. Dialogue-Awakening Mechanism Based on the SOR Theory

By integrating the SOR theory, this study examines the role of conversational activation mechanisms in mitigating driver fatigue (Figure 1). The SOR framework emphasizes the interplay between environmental stimuli (S), internal states of the individual (O), and their behavioral responses (R). Within the context of fatigue driving, specific types of voice-based interactions (stimuli) can influence the driver’s physiological and emotional state (organism), thereby altering their alertness and driving performance (response). Based on our interview data, drivers’ reactions to voice assistant interactions can be categorized into three types: humorous, mechanical/repetitive, and emotionally neutral dialogues. Each type exhibited varying levels of effectiveness depending on the driver’s fatigue state.

Humorous and highly interactive dialogues were found to be the most effective in restoring alertness. These interactions enhanced drivers’ emotional engagement and reduced monotony, especially during prolonged driving. For instance, P01 (long-haul truck driver) noted, “If the assistant tells jokes or asks interesting questions, the journey feels less monotonous and I can stay more alert”. Such interactions were particularly effective for drivers experiencing mild fatigue, helping maintain mental focus and reduce perceived strain.

In contrast, mechanical and low-emotional-value dialogues—characterized by repetitive content and rigid tone—were reported to be ineffective and often counterproductive. P06 (long-haul driver) remarked, “Sometimes the assistant just repeats meaningless reminders, which irritates me and makes the fatigue worse”. These findings suggest that repetitive, non-engaging dialogue lacks the emotional stimulation necessary to sustain driver attention and may even exacerbate fatigue.

The fatigue level emerged as a critical moderating factor. Drivers with mild fatigue responded positively to lighthearted or interactive dialogue. For example, P02 explained, “When I start feeling tired, having light, engaging conversations with the assistant really helps me stay alert”. However, drivers suffering from severe fatigue displayed diminished receptivity—even to humorous dialogue. P09 admitted, “When I’m extremely tired, no joke can wake me up—it just becomes annoying”. In such conditions, cognitive and attentional resources are already depleted, and over-stimulation through dialogue may further increase the cognitive load, leading to aversion.

Therefore, the dialogue content and intervention frequency must be adapted to the driver’s fatigue status. In cases of severe fatigue, the assistant should reduce non-essential interactions and instead provide practical suggestions, such as reminders to take a rest or updates on nearby rest areas. As P03 observed, “When I’ve been driving too long and feel exhausted, a simple rest reminder is enough—no need for extra chatting”.

Under the SOR framework, the effectiveness of conversational stimuli is dependent on the driver’s internal state. Engaging, humorous dialogues can enhance emotional and cognitive arousal, thereby promoting alertness and improving driving performance. P05 (truck driver) confirmed, “Jokes or light conversations help lift my mood and keep me focused”. These reactions support longer periods of safe driving by alleviating subjective fatigue.

Conversely, mechanical dialogues not only fail to produce positive emotional responses but may also intensify feelings of drowsiness or irritability. P04 stated, “When I hear those useless robotic prompts, it just ruins my mood and makes me feel even sleepier”. This reinforces the need for voice assistants to avoid emotionally flat, repetitive content, especially during fatigue-prone periods.

In summary, under the SOR theoretical lens, humorous and interactive dialogues can be effective stimuli to counteract mild fatigue by restoring driver alertness. However, mechanical dialogue tends to exacerbate fatigue regardless of the driver’s condition. Voice assistant systems should dynamically adapt their interaction style and intervention frequency based on real-time assessment of the driver’s fatigue level to optimize responsiveness and minimize fatigue-related risks.

4. Driving Simulation Experiment

4.1. Research Objective

The preliminary interviews, grounded in Grounded Theory and the SOR framework, revealed significant differences in driver preferences for voice assistant stimulation intensity and interaction style under varying fatigue states. However, as the interview study primarily relied on subjective self-reports, it lacked empirical validation at the behavioral level. To enhance both the breadth and depth of the research, this study conducted a driving simulation experiment to analyze the driving performance and subjective experiences of different types of drivers under three voice stimulation conditions. The aim was to examine the adaptability and effectiveness of voice interaction strategies and to validate the findings from the earlier qualitative phase.

4.2. Driving Simulation Experiment Design

4.2.1. Experimental Protocol

This experiment aimed to investigate the awakening effects of different voice interaction styles on driver alertness under varying levels of fatigue. The protocol consisted of two main parts: a fatigue induction phase and an alertness stimulation phase, with a total duration of approximately 50 min. Each participant was required to complete nine experimental sessions, representing all combinations of the three fatigue levels and three types of voice interactions (

Table 4. Experimental design for driver fatigue arousal study.

Fatigue Level\Voice Type	High-Intensity (H)	Moderate-Intensity (M)	Low-Intensity (L)
Non-fatigued state (NFS)	Fast speech rate, high pitch, proactive guidance, with emotional tone, such as ‘Hey, don’t zone out! Let’s guess a riddle to relax!’	Gentle tone, friendly meaning, with a social nature, such as ‘How are you? I can tell you a light joke’	Only necessary driving prompts, no proactive interaction, such as ‘Speed limit ahead 60 km/h, please control your speed’.
Mild fatigue state (MFS)
Severe fatigue state (SFS)

).

Before the experiment began, the driver’s fatigue level was preliminarily assessed using the Karolinska Sleepiness Scale (KSS) to ensure their state met the target fatigue level required for this study. A KSS score of ≤3 was defined as no fatigue, 4 to 6 as mild fatigue, and ≥7 as severe fatigue [31,32]. Considering that fatigue naturally increases with the duration of simulated driving, the experiment was designed to progress sequentially from low to high fatigue levels: awakening trials were first conducted under no fatigue or mild-fatigue conditions, and after each trial, the KSS score was reassessed. Once the next fatigue level was reached, the corresponding awakening trial was carried out.

To effectively induce the target fatigue state, the experiment included a long, monotonous straight-line driving task, requiring participants to drive continuously for 30 min in the simulator. Research shows that driving for 30 min in a monotonous environment can induce significant fatigue [33]. After each driving session, participants completed the KSS to evaluate their fatigue level; if the required fatigue level was reached, they proceeded to the next fatigue-awakening trial; if not, fatigue induction continued until the target state was achieved. To protect participants’ rights, the experimental procedures were fully explained beforehand, informed consent was obtained, participants could withdraw at any time, and appropriate compensation was provided after the experiment.

The experimental driving scenario consisted of a simulated two-way, two-lane road with a total length of 4000 m, including a 3000 m tunnel section. The starting point was located 600 m before the tunnel entrance, and the endpoint was 400 m beyond the tunnel exit. The designed driving speed for the entire segment was 60 km/h. To further evaluate drivers’ vigilance and responsiveness under different voice interaction stimuli, a car-following task was introduced: the lead vehicle traveled at a constant speed of 60 km/h but suddenly decelerated to 30 km/h approximately 2000 m before the tunnel entrance, thereby prompting the participant’s attentional and braking responses to a simulated sudden event.

4.2.2. Simulation Experiment Participants

This study recruited 30 male participants with actual driving experience, aged between 25 and 55 years. All participants had normal vision, were in good physical health, had no history of epilepsy, mental illness, or severe motion sickness, and possessed a valid driver’s license with over three years of driving experience and no record of serious traffic violations. Participation was voluntary, and all individuals provided informed consent prior to the experiment. To analyze how different driver groups respond to voice-based wake-up stimuli under fatigue conditions, participants were categorized into three groups based on their driving occupation: (1) Long-haul truck drivers (n = 10): This group typically engages in prolonged continuous driving and is at a high risk of fatigue. (2) Taxi drivers (n = 10): These drivers face intensive task loads and complex road environments during daily operations, resulting in high cognitive demand. (3) Private car owners (n = 10): This group has a moderate driving frequency, diverse usage scenarios, and greater individual variability.

4.2.3. Experimental Instruments

This experiment utilized three main instruments for data collection and simulation (see Figure 2): First, the Dikablis eye-tracking system (Ergoneers GmbH, Geretsried, Germany) paired with D-Lab analysis software (version 3.55, Ergoneers GmbH, Geretsried, Germany) was employed to capture drivers’ eye movement characteristics. This lightweight and easy-to-use device operates at a sampling frequency of 60 Hz with an accuracy range of 0.1° to 0.3°, allowing real-time monitoring of eyelid closure and pupil dynamics. Second, a 32-channel NE wireless EEG system (Neuroscan, Charlotte, NC, USA) with a high resolution and a sampling rate of 500 SPS was used to accurately record brainwave signals. Various analytical methods were applied to extract power values of α, β, and θ waves as well as EEG waveforms, reflecting drivers’ cerebral activation states. Finally, the DSR-1000TS2.0 driving simulator system (Kunming University of Science and Technology Science & Technology Industrial Management Company, Kunming, China), comprising a cockpit, control console, and three 60-inch 4K LCD monitors, provided an immersive indoor driving environment. It recorded multiple driving behavior parameters such as speed, acceleration, and steering wheel angle to comprehensively assess driving performance.

4.2.4. Experimental Indicators

(1)

Reaction Time: Reaction time refers to the duration between the appearance of an unexpected risk event and the driver initiating a corresponding response, specifically the time from the onset of the hazard to the moment the driver begins pressing the brake pedal. This metric is used to characterize driver alertness [34].

(2)

Arousal: Arousal indicates the participant’s activation state, characterized by high beta activity and low alpha activity consistently observed in the parietal region [35]. The index is calculated as follows: Arousal = α (AF3 + AF4 + F3 + F4)/β (AF3 + AF4 + F3 + F4).

(3)

Eyelid closure degree (PERCLOS): PERCLOS refers to the proportion of time within a unit period that the eyes remain closed. This experiment uses the P80 criterion, which defines the eye as closed when the eyelid covers more than 80% of the eyeball area. This measure reflects the driver’s fatigue level and alertness state [36].

4.3. Experimental Results

4.3.1. Reaction Time

In this study, a driver’s reaction time was defined as the duration from the onset of a sudden braking event of the leading vehicle to the moment the driver initiated braking during the simulated driving task. This measure was used to assess changes in driving alertness under different fatigue states. Reaction time data were collected from 30 participants, and a statistical analysis was conducted to examine how the fatigue level and voice stimulus intensity jointly influence drivers’ responsiveness to sudden hazards (Figure 3). A two-way repeated-measures ANOVA was employed to systematically evaluate the effects of these variables. The results showed a significant main effect of the fatigue level on the reaction time (p < 0.001). As fatigue severity increased, participants’ average reaction time exhibited a marked upward trend, indicating a suppression of driving alertness. This suggests that higher levels of fatigue are associated with delayed recognition and responses to sudden events, thereby posing greater safety risks.

Moreover, the type of voice interaction also had a significant effect on reaction times (p < 0.05). High-stimulation voice prompts—characterized by a faster speech rate, lively intonation, and active engagement—effectively enhanced driver alertness and reduced the reaction time. In contrast, low-stimulation prompts that merely provided passive information had limited arousal effects, resulting in the longest reaction times.

The analysis revealed a significant interaction effect between the fatigue level and type of voice interaction (p < 0.05). Specifically, under severe fatigue conditions, high-stimulation voice prompts reduced the reaction time by approximately 300 ms compared to low-stimulation prompts, demonstrating a marked effect in enhancing alertness. In contrast, when participants were in a non-fatigued state, the differences in reaction time across different types of voice prompts were not statistically significant (p > 0.05). These findings indicate that high-stimulation voice interactions offer stronger driving assistance under higher fatigue levels and can effectively counteract fatigue-induced attentional decline.

Although the main effect of the occupation type was not statistically significant (p > 0.05), preliminary observations suggested trends of differential response patterns among driver groups under specific conditions. Long-haul truck drivers showed the greatest improvement in reaction time under severe fatigue when exposed to high-stimulation prompts, indicating a heightened sensitivity to arousing stimuli. Private car drivers responded most favorably to medium-stimulation prompts under mild fatigue, with notable improvements in reaction time. Taxi drivers exhibited relatively stable responses across different voice interaction types, with medium-stimulation prompts yielding the highest response efficiency.

4.3.2. Arousal

This study evaluated changes in driver alertness by measuring subjective arousal levels during simulated driving tasks under varying fatigue levels and types of voice interaction (Figure 4). Overall, the fatigue level had a significant effect on arousal, with a clear decline in arousal observed as the fatigue severity increased. The voice interaction type also significantly influenced arousal (p < 0.001), with high-stimulation voice prompts yielding the highest average arousal levels. Moreover, a significant interaction effect was found between the fatigue level and voice interaction type (p < 0.05). In particular, under severe fatigue conditions, high-stimulation voice prompts significantly enhanced drivers’ arousal levels, indicating their potential to effectively restore alertness in fatigued drivers. However, under non-fatigued conditions, differences in arousal across voice interaction types were not statistically significant (p > 0.05).

Further analysis of EEG arousal responses among drivers of different occupations revealed that long-haul truck drivers were the most sensitive to arousal enhancement from high-intensity voice stimuli, exhibiting the greatest increase in arousal, especially under severe fatigue conditions. Taxi drivers showed a better response to medium-intensity voice stimuli, while private car owners demonstrated a relatively lower sensitivity to voice stimulation and preferred milder, more gentle interaction styles.

4.3.3. PERCLOS

This study uses the eyelid closure degree during simulated driving—measured by PERCLOS—as a core physiological indicator of fatigue to evaluate changes in alertness under different fatigue states. A higher PERCLOS value typically indicates greater driver fatigue and lower alertness. Based on the PERCLOS data collected from 30 participants across various experimental conditions, a two-way repeated-measures ANOVA was conducted to examine the effects of the fatigue level and voice interaction type on the physiological responses associated with fatigue (Figure 5).

The results show a significant main effect of the fatigue level on the eyelid closure degree (p < 0.001). As fatigue severity increased, PERCLOS significantly rose, indicating that PERCLOS is an effective indicator of the fatigue level. The voice interaction type also exhibited a significant main effect (p < 0.001), with the highest average PERCLOS observed under low-stimulation voice conditions and the lowest under high-stimulation voice conditions, demonstrating that stronger interactive stimuli help alleviate ocular fatigue. Furthermore, a significant interaction effect between the two factors was found (p < 0.05). Notably, under severe fatigue, a high-stimulation voice reduced PERCLOS by approximately 5.2% on average, showing a clear arousal effect; whereas in the no-fatigue state, differences among voice types were not statistically significant.

Additionally, exploratory comparisons of eyelid closure trends among different driver types revealed that long-haul truck drivers experienced the most pronounced PERCLOS reduction under a high-stimulation voice; taxi drivers responded more positively to a medium-stimulation voice; and private car owners showed weaker responses to a high-stimulation voice during mild fatigue, preferring milder interaction styles. Although these differences did not reach statistical significance (p > 0.05), they suggest potential individualized response tendencies in a fatigue intervention across driver groups.

5. Discussion

5.1. The Complex Causes of Fatigue and the Role of Psychological Factors

Fatigue driving is commonly attributed to physiological causes, such as prolonged driving, insufficient rest, or sleep deprivation [37,38,39]. However, with the advancement of research, psychological factors have increasingly been recognized as critical contributors to driving fatigue [40]. While existing studies have predominantly focused on physiological causes, findings from the present study suggest that psychological elements—such as stress, loneliness, and emotional fluctuations—also have a significant impact on the level of fatigue. These factors can exacerbate drivers’ fatigue states, particularly during long-haul or monotonous driving conditions, underscoring the importance of a more comprehensive approach to fatigue prevention.

Existing research has primarily focused on the physiological causes of driver fatigue. Prolonged driving durations and sleep deprivation have been consistently identified as the dominant physiological contributors. Additionally, monotonous driving environments—such as long stretches of straight roads—are known to accelerate the onset of fatigue by reducing cognitive stimulation and attentional engagement [41]. However, these studies often overlook the role of psychological factors. In contrast, the present study highlights the significance of psychological factors in fatigue driving, especially during extended driving periods or night-time driving. Psychological stress and emotional fluctuations can exacerbate fatigue. The research finds that when drivers engage in long periods of solitary driving or face high-pressure transport tasks, emotional instability and stress significantly impact their cognition and attention, thereby intensifying the fatigue state.

Fatigue detection has predominantly focused on assessing driver fatigue through physiological signals, such as heart rate and eye movements [42], often overlooking the influence of psychological factors. This study reveals that psychological factors, such as loneliness and emotional fluctuations, especially during night-time driving or long-duration transportation, significantly increase drivers’ fatigue and may exacerbate cognitive fatigue and attention decline. This contrasts with traditional research, which has been dominated by physiological factors, offering a more comprehensive perspective.

Furthermore, research indicates that psychological stress not only affects drivers’ emotional states but also alters their coping strategies and response capabilities to driving tasks [43]. These effects are particularly pronounced during long periods of driving, making it difficult for drivers to effectively manage the cognitive burden caused by fatigue. Therefore, in contrast to traditional studies that focus solely on physiological fatigue, this research emphasizes the fatigue states of drivers under high pressure and emotional fluctuations, thereby expanding the understanding of fatigue’s underlying causes.

5.2. The Impact of Occupation Type on Dialogue Preferences

The occupational background largely shapes drivers’ dialogue needs and preferences in the context of fatigue driving. This difference stems not only from the nature of the driving tasks themselves but is also closely related to the long-term work pressure, social environment, and contextual factors experienced by the drivers. This study found that the occupational type significantly influences drivers’ interaction patterns with voice assistants, reflected in differences in dialogue content, functional demands, and interaction styles. However, current research in this field remains limited, particularly regarding the exploration of the underlying psychological mechanisms behind these occupational differences.

For long-haul freight drivers, their work is characterized by prolonged, monotonous, and high-intensity driving tasks, often accompanied by significant time pressure and feelings of loneliness [44]. This high cognitive load and low-social-interaction driving context make them more inclined to task-oriented voice interactions when fatigued, such as navigation, traffic information alerts, and fatigue warnings. These functional dialogues not only fulfill their need for information but also help maintain attention and reduce feelings of fatigue. However, this study further found that after driving continuously for a certain period—especially during nighttime—these drivers also exhibit a need for light social interactions, such as simple Q&A or casual chatting, which may serve as a coping strategy against prolonged social isolation and cognitive depletion.

Taxi and rideshare drivers, on the other hand, operate in a different typical context. They frequently navigate complex urban traffic environments and engage in frequent interactions with passengers. Although this occupational group also faces serious fatigue issues [45], their social load is relatively higher, which shapes their preference for voice interactions that emphasize social and emotional regulation functions. Research shows that this group is more receptive to voice content with an affiliative or social nature, such as casual weather chats, friendly greetings, or discussions about daily life, and they even hope that voice assistants can simulate a “companionship-style” dialogue. This demand reflects their psychological need for emotional buffering after intense social work, with voice interactions serving as an alternative form of social engagement that helps release emotions and alleviate fatigue.

In contrast, private car owners exhibit more diverse driving behaviors and contexts, and their reliance on voice assistants is relatively lower. Most private car owners have limited daily driving time, typically for commuting, shopping, or leisure trips, with work-related stress mainly stemming from balancing their life pace and family roles [46]. In such relatively relaxed driving scenarios, voice interactions tend to be more entertainment- and emotion-support oriented, such as playing music, delivering news broadcasts, or providing voice reminders related to family communication. They prefer low-intensity, non-task-oriented dialogue content to alleviate commuting fatigue and create a relaxed atmosphere.

5.3. The Role of Interactive Dialogue in Fatigue Recovery

In recent years, the importance of interactive dialogue in alleviating driving fatigue and enhancing driver alertness has garnered increasing attention. This study finds that highly interactive voice dialogues, particularly humorous and Q&A-style dialogues, have a significant effect on fatigue recovery. This conclusion aligns with existing research, supporting the notion that interactive dialogue can activate drivers’ emotions, improve their mental state, and enhance their attention [47]. Compared to mechanical, one-way information delivery, dialogues with emotional and cognitive interactions are more likely to trigger positive emotions in drivers, alleviate negative emotions and psychological stress, and thus effectively counter the effects of fatigue.

In long-duration driving tasks, such as long-haul freight, taxi, or ride-hailing operations, drivers demonstrate a higher demand for voice dialogues that are both interactive and emotionally valuable. When the emotional expression of the voice system aligns with the driver’s emotional state, drivers are less likely to have accidents, pay more attention to road conditions, and are more likely to engage in interactions [48]. Q&A-style dialogues are particularly common in these scenarios, fulfilling the need for information while also helping to divert attention and restore cognitive alertness. Additionally, voice interaction systems have an advantage over touch-based systems in reducing distractions and improving accessibility [49].

However, the positive effects of interactive dialogue are not universally effective in all contexts. On the one hand, individual differences significantly impact the interactive experience. For instance, older drivers or those with a higher dependence on technology may tend to focus more on interface feedback during voice interactions, which can pose safety risks [50]. On the other hand, in cases of severe fatigue, excessive interaction or information input may increase the driver’s cognitive load and exacerbate fatigue. Therefore, the content and frequency of interactions should be tailored to the driver’s fatigue state: light, humorous dialogue or casual chats can aid fatigue recovery in mild fatigue, while unnecessary interventions should be minimized in severe fatigue to avoid cognitive overload [51].

Furthermore, this study reveals that drivers from different occupational backgrounds and with varying individual characteristics have different preferences for voice interactions. Long-haul freight drivers tend to prefer a balance of functionality and interactivity in voice dialogues to alleviate fatigue from long hours of driving. In contrast, taxi drivers and private car owners are more inclined towards social and entertainment-oriented interactions. This finding confirms that human drivers exhibit specific social preference patterns during driving tasks [52].

5.4. The Necessity of Adaptive Voice Interaction

With the continuous advancement of intelligent voice assistants and autonomous driving technology, the importance of adaptive voice interaction in reducing driving fatigue and enhancing driving safety has become increasingly prominent [53]. Based on SOR theory, a driver’s fatigue state significantly influences their behavioral responses, and the interaction strategy of a voice assistant needs to dynamically adjust based on the driver’s fatigue level, contextual demands, and individual preferences. The findings of this study show that when the interaction content and methods of the voice assistant can flexibly adapt to the driver’s changing state, they are more effective in helping alleviate fatigue, improve alertness, and ultimately enhance driving safety.

In severe fatigue states, a driver’s cognitive function and emotional stability significantly decline, with slower reaction times and distracted attention [54]. At this point, the voice assistant should adopt a high-intensity dialogue strategy, such as humorous anecdotes or Q&A-style interactions, to provide emotional stimulation and cognitive awakening, quickly capture attention, and restore alertness. In contrast, mechanical, repetitive, or emotionally detached voice content often fails to have a positive effect and may even worsen the driver’s fatigue, leading to negative emotional responses and cognitive breakdown. This further validates that interacting with the system helps significantly reduce driver fatigue and boredom [55].

For mild fatigue states, although drivers exhibit initial signs of fatigue, their cognitive abilities remain relatively stable. In this situation, the voice assistant can use low-intensity, non-task-oriented, social dialogues, such as casual chats or informal greetings, to help the driver release stress, maintain a positive mood, and continue focusing on the road environment. This type of interaction reduces psychological load during driving while avoiding cognitive interference, helping delay fatigue development and enhancing driving continuity [56]. Studies have also shown that voice robots can enhance drivers’ sense of safety and calmness through engagement and emotional regulation during simulated driving [57] and improve alertness in autonomous driving through active listening and voice reminders [58].

In a non-fatigued state, the voice assistant’s intervention should be minimized. Excessive voice prompts or irrelevant interactions may interrupt the driver’s focus on the current traffic environment, leading to distraction or cognitive load, thereby affecting driving performance. Therefore, the voice assistant should avoid unnecessary interventions unless the driver initiates communication. In this state, voice interaction should be task-oriented and concise, assisting the driver in focusing on the driving task [59].

Beyond fatigue states, a driver’s personalized needs are also an important variable in designing voice assistant interaction strategies. Factors such as the voice assistant’s voice characteristics, tone, gender, and privacy protection mechanisms significantly influence its acceptance [60]. Different occupational groups exhibit varying driving behaviors and interaction needs, requiring differentiated designs. For instance, long-haul freight drivers prefer task-oriented voice support, such as real-time traffic information and navigation prompts; taxi drivers are more sensitive to passenger-related interactions and may prioritize the voice assistant’s social and contextual awareness; while private car owners tend to prefer light, entertainment-oriented dialogues to alleviate driving pressure and fatigue. These differences suggest that future voice interaction systems should enhance their ability to recognize and respond to the emotional states and personalized needs of drivers, achieving truly dynamic adaptation and intelligent collaboration, thus ensuring safety while improving the user experience.

5.5. The Role of Voice Stimuli in Enhancing Alertness Under Fatigue

This study, through a simulated driving experiment, confirmed the significant inhibitory effect of fatigue on driver alertness, as evidenced by prolonged reaction times, decreased subjective arousal, and increased the eyelid closure rate. These findings are consistent with a large body of previous research [39,61,62], further underscoring the negative impact of fatigue on driving safety. As a non-intrusive intervention method, voice interaction—particularly high-stimulation voice stimuli—significantly improved alertness under conditions of severe fatigue. A real-world driving study also found that engaging drivers in conversation noticeably enhanced their alertness [63].

Moreover, this study revealed a significant interaction between the fatigue level and voice stimulus intensity, indicating that the effectiveness of voice stimuli depends on the driver’s current fatigue state. This finding aligns with the view that the effectiveness of fatigue mitigation interventions varies with fatigue severity, highlighting the importance of dynamically adjusting the intervention’s intensity [64,65]. Differences in voice stimulus responsiveness among professional driver groups suggest that future intelligent driver assistance systems should incorporate drivers’ occupational backgrounds and personal preferences to achieve more targeted and effective interventions. For instance, private car drivers showed weaker responses to high-stimulation voice stimuli under mild fatigue but expressed a stronger preference for personalized interactions, suggesting that the content and style of voice stimuli should be customized—a key direction for future research.

5.6. Suggestions

Based on the systematic analysis of fatigue causes, occupational differences, interactive dialogue effects, and voice assistant adaptability in this study, the following recommendations are proposed to provide theoretical support and practical guidance for the design and application of intelligent voice interaction systems, while enriching the knowledge framework in this field:

(1)

Incorporate Psychological Factors into Fatigue Detection Models

Current mainstream fatigue detection primarily relies on physiological indicators such as the heart rate, eye movement, and posture changes. However, this study reveals that psychological states—including emotional fluctuations, feelings of loneliness, and stress—have a significant impact on fatigue formation. Integrating psychological dimensions such as emotion monitoring, speech emotion recognition, and tone analysis into multimodal fatigue detection can not only improve detection accuracy but also provide a basis for personalized intervention, thereby enriching the theoretical framework of fatigue detection. This direction offers new entry points for future research and facilitates the transition of fatigue management technologies from relying solely on physiological parameters to a comprehensive integration of psychological and physiological factors.

(2)

Implement Personalized Voice Interaction Design Based on Occupational Types

This study systematically reveals for the first time the significant differences in dialogue needs among drivers of different occupational backgrounds, highlighting the profound influence of the occupation on voice interaction preferences. Intelligent voice systems should tailor interaction strategies according to users’ occupational characteristics: emphasizing functional, task-oriented Q&A modes for long-haul freight drivers; focusing on social and emotional regulation dialogue content for taxi and rideshare drivers; and designing relaxed, entertainment-oriented, low-intensity interactions for private car owners. This design concept not only supplements the understanding of user diversity in intelligent driving human–machine interaction but also provides clear pathways for developers to implement differentiated customization.

(3)

Develop a Dynamic Adaptation Mechanism for Fatigue Dialogue Intervention

Drivers’ acceptance of voice interaction varies significantly with fatigue levels. By integrating real-time fatigue detection, dynamically adjusting dialogue intensity and content can achieve precise and effective vocal arousal and psychological relief. This mechanism overcomes the static limitations of traditional fatigue management and promotes the system’s shift from passive detection to active intervention. Future research may explore more intelligent interaction strategies based on this framework, such as deep integration of multimodal sensor fusion and context awareness, thereby enhancing the safety assurance capabilities of intelligent driving systems.

(4)

Enhance Contextual Awareness and Adaptation to Individual Differences

Fatigue manifestations are influenced with multiple factors, including the driving environment, time cycles, and task pressure. Future intelligent voice assistants should combine context-aware technologies—such as time, geographic location, and task type sensing—with user historical behavior and emotional preference modeling to realize more precise and personalized dialogue strategy adjustments. This not only enriches the adaptability theory of voice interaction but also offers a technical route for optimizing the user experience in intelligent driving systems, contributing to improved user retention and application effectiveness.

(5)

Shift Fatigue Management from Passive Detection to Active Intervention

Traditional fatigue management largely depends on post-event reminders, which have limited intervention timeliness. This study demonstrates that moderate voice interactions can provide psychological relief and restore attention at early stages of fatigue, effectively enhancing driving safety. It is recommended that future systems strengthen emotion computing and behavior prediction capabilities to build a closed-loop management system of “perception–prediction–intervention–evaluation”, achieving full-process active intervention. This not only enriches the theoretical system of fatigue management but also provides practical pathways for innovation in safety strategies within intelligent driving systems.

5.7. Limitations

First, the sample size was small and geographically concentrated in Fujian Province, China, which may limit the generalizability of this study’s findings. Cultural and regional differences could significantly influence driving behaviors and preferences for voice assistant interactions. For example, users’ expectations regarding tone of voice, interaction frequency, or emotional expression may vary based on cultural backgrounds. Therefore, the findings of this study should be interpreted with caution, taking participants’ cultural contexts into account. Future research should aim to expand the geographical and cultural diversity of the sample to enhance the external validity of the results. In addition, although the voice interaction strategies were validated in a driving simulation environment, the experimental scale was limited, and the personalization mechanisms remain at an exploratory stage. Further empirical validation is needed in future work, ideally incorporating physiological monitoring in real-world driving contexts to optimize differentiated interaction strategies and strengthen the scientific rigor and practical applicability of the proposed dialogue-awakening mechanism.

6. Conclusions

By integrating the SOR theoretical framework with the Grounded Theory methodology, this study conducted in-depth interviews with 25 drivers from different occupational backgrounds to systematically investigate their preferences for voice assistant interaction content, format, and frequency under varying levels of driving fatigue. The findings reveal that drivers’ interaction demands are significantly influenced by their fatigue state. Specifically, under severe fatigue, most drivers preferred highly stimulating voice prompts—such as rhythmic question-and-answer formats or humorous motivational messages—to rapidly enhance alertness. During mild fatigue, they tended to favor more soothing and gentle conversational styles, such as weather updates or music recommendations. In a non-fatigued state, drivers generally preferred a quiet driving environment and only engaged with voice interactions when necessary. Clear differences were also observed across driver groups: long-haul truck drivers prioritized practicality and frequent wakefulness prompts; taxi drivers favored interactive modes integrating navigation and social features; while private car owners showed a stronger preference for personalized and emotionally supportive interactions.

To validate the qualitative findings, a driving simulation experiment was conducted to empirically examine the effect of different levels of voice stimulus intensity on driver alertness across varying fatigue levels. The experimental results supported the interview-based conclusions, particularly highlighting that high-intensity voice interactions significantly improved alertness under severe fatigue. This reinforces the practical value of voice interaction as an effective fatigue intervention strategy.

Overall, this study not only extends the application of the SOR framework to the field of intelligent voice interaction but also provides empirical evidence for designing personalized voice assistants based on drivers’ fatigue states and occupational characteristics. This study highlights the behavioral adaptability potential of voice interaction systems in traffic safety interventions, promoting a shift from purely passive fatigue detection to an active human–machine collaborative intervention model. This approach holds significant social safety value and broad prospects for widespread application. Future research can further integrate multimodal sensing technologies to deepen the understanding of contextual factors and cultural differences, thereby enhancing the safety and user experience of intelligent driving systems and advancing their innovation and widespread adoption.