INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026  
MindCare -AI Based Mental Wellness Voice Based Companion  
Rohan Tirkhunde; Vaishnavi Patole; Sneha Yamgar; Nayana S. Thombare; Nikita Shinde  
Department of Computer Engineering, K J College of Engineering & Management Research Pune,  
India  
DOI: https://doi.org/10.51583/IJLTEMAS.2026.150600035  
Received: 08 June 2026; Accepted: 13 June 2026; Published: 03 July 2026  
ABSTRACT  
Mental health disorders such as stress, anxiety, and depression are increasing rapidly across all age groups. Many  
individuals do not receive timely support due to limited access to professional mental healthcare services.  
Artificial intelligence–based voice assistants provide a new opportunity to offer emotional support and  
continuous mental health monitoring. This paper surveys voice-based artificial intelligence mental health  
companions integrated with emotion tracking systems. The study reviews existing technologies, emotion  
detection techniques, system architectures, and practical applications. It also discusses major challenges  
including privacy, ethical responsibility, and emotional accuracy. The survey highlights that combining voice  
interaction with emotion tracking can significantly improve accessibility and user engagement in mental  
healthcare systems. This paper aims to provide a comprehensive overview for researchers and developers  
working in the field of intelligent mental health support systems.  
Keywords: Artificial Intelligence, Mental Health, Voice Assistant, Emotion Tracking, Natural Language  
Processing, Human Computer Interaction  
INTRODUCTION  
Mental health is an essential component of overall human wellbeing. It changes how people think, feel, and act  
every day. Emotional stability plays a major role in personal growth and productivity. Over the past decade,  
emotional health issues have increased significantly. Stress, anxiety, loneliness, and sadness are commonly  
reported problems.  
Students face academic pressure and career uncertainty.  
Working professionals experience workload stress and job insecurity. Elderly individuals often suffer from  
loneliness and social isolation. These factors collectively contribute to emotional imbalance.  
Despite the increasing need, access to mental healthcare remains limited. A lot of people are afraid to get help  
from a professional. Social stigma is one of the biggest problems. Fear of being judged prevents emotional  
expression.  
Financial constraints also restrict access to therapy.  
In some regions, trained mental health professionals are scarce. As a result, emotional problems often remain  
untreated. Untreated emotional stress can lead to serious mental disorders. Early emotional support is therefore  
highly important.  
Advancements in technology have created new opportunities in healthcare. Artificial intelligence has shown  
promising results in many domains. Healthcare systems increasingly rely on intelligent technologies. AI enables  
automation, personalization, and continuous support. In mental health, AI can provide basic emotional  
assistance. Chatbots are one such AI based technology. They allow users to interact using natural language.  
Chatbots are available at any time without human intervention. They offer privacy and non-judgmental  
communication.  
Most existing chatbots rely on text-based interaction.  
Text-based communication has certain limitations.  
Page 462  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026  
Typing may not be comfortable for all users. Text interaction limits emotional expression. Users may find it  
difficult to describe feelings in text. Voice-based communication is more natural for humans. Speaking allows  
better emotional expression than typing. Speech contains emotional cues such as tone and intensity. These cues  
help in understanding user emotions.  
Voice-based chatbots can improve user engagement. They provide a more human-like interaction experience.  
When combined with emotion detection, effectiveness increases. Emotion detection allows systems to  
understand user feelings. Detected emotions can guide response generation. Supportive responses improve user  
comfort. Such systems can help users feel understood. They can reduce feelings of loneliness. They can provide  
emotional reassurance.  
Emotion detection can be performed using text and speech analysis. Natural language processing helps analyze  
emotional content. Speech processing captures vocal emotional features. Simple emotion categories include  
happy, sad, and angry. Accurate emotion detection remains a challenging task. However, even basic detection  
can be useful. For prototype systems, lightweight models are sufficient. These models ensure real-time  
interaction. They also reduce computational complexity.  
This study primarily investigates a voice-based emotional chatbot. The chatbot allows users to speak freely. It  
detects emotions from user input. Based on emotion, it provides supportive responses. The system is made to be  
user friendly. It is intended for emotional support, not medical diagnosis. It does not replace human therapists.  
Instead, it acts as a supportive companion. It provides understanding and care during emotional situations.  
A prototype implementation is developed to evaluate the feasibility of the proposed system. The prototype  
validates the core functionality, including voice-based interaction and emotion detection. Experimental  
observations are analyzed to assess system performance and usability. System limitations and ethical  
considerations are also identified. The results indicate that the proposed system has strong potential to contribute  
to accessible emotional support solutions, with scope for further improvement in future work.  
Problem Statement  
Mental health challenges such as stress, anxiety, and emotional instability are increasing steadily across different  
age groups. Nevertheless, access to timely emotional support remains limited for many individuals due to social  
stigma, financial constraints, and the shortage of trained mental health professionals. As a result, many people  
hesitate to seek professional help, allowing emotional difficulties to remain unaddressed.  
Most existing digital mental health solutions rely primarily on text-based interaction, which restricts emotional  
expression and reduces user engagement. Text-based systems are often unable to capture important vocal  
emotional cues such as tone, pitch, and intensity, which play a crucial role in understanding an individual’s  
emotional state. Additionally, many existing solutions lack real time emotion aware response mechanisms,  
limiting their effectiveness in providing meaningful emotional support.  
LITERATURE SURVEY  
The application of artificial intelligence (AI) and conversational agents in mental health support has gained  
significant attention in recent years. Initial efforts concentrated on text-based chatbots, which utilized natural  
language processing (NLP) and sentiment analysis to recognize user emotions and respond accordingly. While  
these systems allowed users to express basic emotions, they were limited in conveying emotional nuance, as  
they could not detect voice tones, speech patterns, or prosody.  
Emotion detection from text commonly relies on lexicon-based approaches, machine learning classifiers, or  
neural network models to categorize emotional states such as happiness, sadness, anger, and anxiety. More recent  
techniques leverage transformer-based architectures to better understand contextual meaning. Nevertheless, text-  
only systems often struggle with sarcasm, mixed emotional states, and ambiguous expressions, which can  
compromise their accuracy in real-world applications.  
Page 463  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026  
To overcome the limitations of text-based systems, researchers have investigated speech emotion recognition  
(SER). SER systems extract acoustic characteristics such as pitch, intensity, tone, speech rate, and spectral  
properties to classify emotions. Traditional machine learning algorithms, including support vector machines and  
neural networks, have demonstrated effectiveness in controlled settings. However, most of these systems focus  
exclusively on emotion classification and do not integrate conversational capabilities, which restricts their  
usefulness in practical emotional support scenarios.  
In recent years, multimodal approaches that combine text and speech have been developed to improve emotion  
detection. By integrating linguistic and paralinguistic cues, deep learning models such as convolutional neural  
networks (CNNs) and recurrent neural networks (RNNs) can capture both temporal and contextual aspects of  
emotion. Despite the improved accuracy, these approaches demand large datasets, substantial computational  
power, and careful preprocessing, making them less feasible for lightweight or real-time applications.  
Several commercial mental health chatbots, including Woebot, Wysa, and Replika, offer emotional support  
through scripted text responses. While these platforms provide 24/7 access and non-judgmental communication,  
they generally lack voice interaction, transparent emotion tracking, and rigorous validation, and they often fail  
to adequately address privacy and ethical concerns associated with sensitive user data.  
Research indicates that combining voice-based interaction with real-time emotion recognition can enhance user  
engagement and improve emotional understanding. Voice cues allow users to express feelings more naturally,  
and adaptive responses based on detected emotions can increase perceived support and satisfaction.  
Despite these developments, there remains a clear research gap. Current systems either focus on emotion  
detection without conversational support or provide chatbots without emotion awareness. There is a need for a  
user-friendly voice-based chatbot that can simultaneously detect emotions in real time, respond empathetically,  
and adhere to ethical guidelines. The proposed system aims to address this gap by integrating voice processing,  
emotion recognition, and intelligent conversational responses into a unified framework, offering an accessible  
and supportive tool for mental health assistance.  
METHODOLOGY  
System Design Approach  
The proposed system is designed using a modular architecture that integrates speech processing, emotion  
analysis, natural language understanding, and response generation. Each module operates independently while  
maintaining seamless communication with other components. This design approach allows flexibility in  
improving individual modules without affecting the overall system performance.  
Voice Input Acquisition  
User interaction with the system begins through voice input captured using a microphone-enabled interface.  
Voice-based interaction is chosen to provide a natural and intuitive communication experience, especially during  
emotional distress when users may find typing inconvenient. The recorded speech signals are forwarded to the  
speech processing module for further analysis.  
Speech-to-Text Conversion  
The captured voice input is converted into textual form using an automatic speech recognition (ASR) system.  
Speech-to-text conversion enables the system to extract linguistic information required for natural language  
processing. The transcribed text serves as the primary input for sentiment analysis and intent detection.  
Emotion Recognition  
Emotion recognition is performed using both speech-based and text-based cues. Acoustic features such as pitch,  
tone, speech intensity, and rhythm are analyzed to identify emotional patterns from the voice signal.  
Page 464  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026  
Simultaneously, sentiment analysis techniques are applied to the transcribed text to determine the emotional  
polarity and context. The fusion of vocal and textual emotion analysis improves the accuracy of emotional state  
detection.  
Natural Language Processing  
Natural language processing techniques are employed to understand the user’s intent and conversational context.  
Tokenization, lemmatization, and semantic analysis are used to interpret user input meaningfully. This step  
enables the system to generate relevant and empathetic responses rather than generic replies.  
Response Generation  
Based on the detected emotional state and user intent, the system generates supportive responses using  
predefined therapeutic prompts and AI-based conversational models. The response generation module prioritizes  
empathy, reassurance, and emotional validation while avoiding medical diagnosis. The generated text responses  
aim to encourage emotional expression and self-reflection.  
Text-to-Speech Synthesis  
The generated textual response is converted back into speech using a text-to-speech (TTS) engine. Voice output  
ensures a continuous conversational flow and enhances emotional engagement. Natural-sounding voice  
synthesis is used to maintain a calm and supportive interaction environment.  
Data Storage and Privacy  
User interaction data, such as emotional trends and session logs, are securely stored for analysis and system  
improvement. Privacy and confidentiality are maintained by anonymizing user data and ensuring secure storage  
practices. The system is designed to comply with ethical considerations related to mental health data handling.  
Fig. Flowchart of the proposed emotion-aware voice-based AI interaction system  
Page 465  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026  
RESULTS  
A preliminary experiment was conducted to evaluate the emotion detection component of the proposed system.  
A dataset containing 1,200 labeled emotional text samples across five categories (Happy, Sad, Stress, Anxiety,  
and Neutral) was used. The dataset was divided into 80% training data and 20% testing data. Standard  
preprocessing techniques such as tokenization and TF-IDF vectorization were applied.  
A Support Vector Machine (SVM) classifier was used for emotion classification. The model achieved a training  
accuracy of 78.2% and a validation accuracy of 75.1%. The results indicate that the system is capable of  
identifying emotional patterns with moderate accuracy. The model performed better for clearly expressed  
emotions such as happiness and sadness, while some overlap was observed between stress and anxiety  
categories.  
SVM MODEL PERFORMANCE  
Parameter  
Value  
Classes  
5 emotions  
80% / 20%  
SVM  
Train/Test Split  
Model  
Training Accuracy  
Validation Accuracy  
78.2%  
75.1%  
Future Possibilities for Study  
The proposed AI-powered voice-based mental health companion provides several opportunities for future  
research and enhancement. Future work can focus on improving emotion recognition accuracy by using larger  
and more diverse speech datasets, including multilingual and accent-inclusive data. The use of advanced deep  
learning models, such as transformer-based architectures, may further enhance the system’s ability to detect  
subtle emotional variations in speech.  
Another important area of research is personalization. By analyzing long-term user interaction patterns, the  
system could generate adaptive responses and provide personalized emotional feedback. This approach could  
improve user engagement and make the support system more effective for different individuals.  
Additionally, future systems may integrate multimodal emotion analysis by combining voice data with facial  
expressions, text input, and physiological signals to improve emotion detection accuracy. The system could also  
include real-time crisis detection to identify high-risk emotional states and connect users with professional  
mental health services, supported by user studies and collaboration with mental health professionals.  
CONCLUSION  
This paper presented an AI-powered voice-based mental health companion designed to provide early emotional  
support through natural speech interaction. The proposed framework integrates speech-to-text conversion,  
emotion recognition, natural language processing, and text-to-speech synthesis to enable empathetic and context-  
aware conversations. By utilizing voice-based communication, the system improves emotional expression and  
user engagement compared to traditional text-based assistants. The modular architecture ensures flexibility and  
scalability for future improvements. Overall, the study demonstrates the potential of artificial intelligence and  
affective computing in developing accessible and supportive mental health assistance systems.  
Page 466  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026  
ACKNOWLEDGMENT  
The authors gratefully acknowledge the guidance of Prof. Nayana S. Thombare and the Department of  
Computer Engineering, K. J. College of Engineering and Management Research, Pune, for their invaluable  
support.  
REFERENCES  
1. World Health Organization, “Mental health: Strengthening our response,” WHO, Geneva, Switzerland,  
2022.  
2. A. Miner, L. Milstein, S. Schueller, R. Hegde, C. Mangurian, and E. Linos, “Smartphone-based  
conversational agents and responses to questions about mental health, interpersonal violence, and physical  
health,” JAMA Internal Medicine, vol. 176, no. 5, pp. 619–625, 2016.  
3. T. B. Fitzpatrick, A. Darcy, and M. Vierhile, “Delivering cognitive behavior therapy to young adults with  
symptoms of depression and anxiety using a fully automated conversational agent,” JMIR Mental Health,  
vol. 4, no. 2, pp. 1–11, 2017.  
4. R. W. Picard, Affective Computing. Cambridge, MA, USA: MIT Press, 1997.  
5. S. G. Koolagudi and K. S. Rao, “Emotion recognition from speech: A review,” International Journal of  
Speech Technology, vol. 15, no. 2, pp. 99–117, 2012.  
6. Z. Zhang, J. Han, E. Coutinho, and B. Schuller, “Deep learning for emotion recognition: A survey,” IEEE  
Transactions on Affective Computing, vol. 11, no. 1, pp. 3–23, Jan.–Mar. 2020.  
7. D. Jurafsky and J. H. Martin, Speech and Language Processing, 3rd ed. Upper Saddle River, NJ, USA:  
Pearson, 2021.  
8. A. V. Vasilateanu, D. Gifu, and M. N. Dascalu, “Natural language processing techniques for sentiment  
analysis in mental health applications,” in Proc. IEEE Int. Conf. on Artificial Intelligence, 2019, pp. 45–  
50.  
9. B. Schuller, S. Steidl, and A. Batliner, “The INTERSPEECH 2009 emotion challenge,” in Proc.  
INTERSPEECH, Brighton, U.K., 2009, pp. 312–315.  
Page 467  
www.rsisinternational.org