Using Voice to Detect Depression
Author Information
Author(s): Huang Xiangsheng, Wang Fang, Gao Yuan, Liao Yilong, Zhang Wenjing, Zhang Li, Xu Zhenrong
Primary Institution: School of Biomedical Engineering, South-Central Minzu University
Hypothesis
Can a voice-based pre-training model effectively identify depression?
Conclusion
The proposed model achieved high accuracy in detecting depression using voice data.
Supporting Evidence
- The model achieved an accuracy of 0.9649 in binary classification.
- In multi-classification, the model reached an accuracy of 0.9481.
- The study utilized a dataset of 6545 audio samples for training and testing.
Takeaway
This study shows that we can use people's voices to help doctors find out if someone is depressed, making it easier to help them.
Methodology
The study used the wav2vec 2.0 model to extract voice features from audio data and fine-tuned it on the DAIC-WOZ dataset for depression classification.
Potential Biases
The controlled environment of the dataset may not reflect real-world scenarios.
Limitations
The dataset was limited in size and diversity, which may affect the model's generalizability.
Participant Demographics
The dataset included 189 audio files from various age groups, with 102 male and 87 female participants.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website