PSAAP: Patient-Specific Audio Extraction Pipeline for Depression-Oriented Vocal Biomarker Analysis
Raminder Kaur Nagra, Vikram Kulkarni · International Research Journal of Multidisciplinary Technovation · 2025
The stigma of depression and mental illness is growing everywhere in the world and this is the reason why the moves are on to come up with diagnostic tools, which would be rapid, efficient and consistent. The framework suggested in this study is a state-of-the-art technic to infer patient-specific data on therapy-patient dialogs to fill a knowledge gap that existing diagnosis solutions are unable to cover. The proposed signal processing scheme, Patient-Specific Audio Extraction Pipeline (PSAAP) enhances input to the Machine Learning (ML) of mental illness detection. The method locates and measures non verbal acoustic features such as the pitch, the intensity, and the Mel Frequency Cepstral Coefficients (MFCCs) that play a crucial role in the determination of mental good health. Noise reduction, speaker diarization, silence deletion as well as other samples of preprocessing are implemented on DAIC-WOZ dataset to maintain quality of audio and sound. In this code, speech characteristics which are instrumental in such diagnosis have been retained so that the depression symptom described by monotony, slow speech and low variability of pitch can be analyzed precisely. The quantitative findings are demonstrated in the framework and signal-to-noise ratio (SNR) is augmented by up to 16 dB compared to the existing methods. The method allowing the practitioner to make a sound and clinically pertinent evaluation since separating patient-specific variables of the voice and removing therapist feedback makes assessment trait-centered and dependable. The main argument in favour of this one is that it possesses the quality to be applied in the extraction of patient speech framework in general since the patient can find such opportunity to be helpful in the mental healthcare examination and also in the circumstances where the computing capabilities are poor.