Depression often hides in plain sight. At the Lab we are developing tools that listen for the subtle acoustic changes that accompany it, with the aim of making screening timely, objective and accessible.
Every utterance carries two layers: what is said, and how it is said. The second is where we look.
Technology should extend clinical judgement, not stand in for it. Speech is one of the few clinical signals a person produces continuously, at no cost, without instrumentation which makes it worth understanding properly.
Word choice, semantic coherence, narrative structure. Well studied, but bound to language and culture, and dependent on transcription.
Rhythm, prosody, pitch variability, pause structure, voice quality. Largely language-independent and measurable directly from the waveform. This is the layer the lab investigates.
Whether an acoustic signal can support depression assessment depends on holding it up against both clinical populations and the wider community it would eventually serve.
Depression assessment and assistive diagnostic technology sit at the core of the lab. Working under the guidance of investigators in the Department of Psychiatry, and with funding support from Coal India Limited, we develop and refine assessment tools through direct interaction with people living with depression.
Mental health research has to reach past the laboratory. Any diagnostic model needs robust, unbiased baseline data from people who are not patients, across age groups and generations. Working with communities on the foundations of stress and anxiety is how that baseline gets built, and how methods find their way onto web and mobile platforms where people can actually use them.
Each stage is designed to be auditable, so a result can be traced back to the audio it came from.
Consented, ethically approved collection of speech samples under controlled conditions.
Noise reduction, segmentation and normalisation, so recordings made in different rooms remain comparable.
Paralinguistic feature extraction prosody, timing, pause structure and voice quality.
Statistical and machine learning models relating acoustic features to clinical measures.
Cross-reference against structured clinical assessment, with held-out data and clinician review.
Dr Himanshu Singh works full time at the Speech Health Lab as Scientist-C. The work described on this page is carried out under Professor Dr Nand Kumar, Department of Psychiatry, All India Institute of Medical Sciences, New Delhi, and is actively funded by Coal India Limited.
The lab maintains its own site at speechhealth.org. The work is presented here to document the projects Dr Singh contributes to; the lab and its outputs belong to the Department.
Research in progress. Methods described here are under active development and clinical validation. Nothing on this page is a diagnostic tool, and none of it should be used in place of assessment by a qualified clinician.
For collaboration, participation or questions about the methods, get in touch directly.
© 2026 SignalsLab · Himanshu Singh · Data sources & licences