Mobius Labs GmBH

Senior Researcher, Speech Technology December 2020 - Present

• Responsible for audio R&D at Mobius.

• Design, develop, and support product team for all audio-related features at Mobius labs.

• Develop and improve scalable SOTA speech recognition system from videos.

• Custom audio event detection (patent in progress) and text analyses for video understanding.

• Multimodal learning, data-centric AI and model optimization

Telepathy Labs GmBH

Research Engineer, Speech Technology December 2018 - Nov 2020 Present

• Developed state-of-the-art fusion techniques for multi-modal emotion recognition combining speech and transcripts.

• Implementing emotion recognition and sentiment analysis services from speech and text.

Idiap Research Institute

Pre-doctoral fellow September 2017 – August 2018

• Direct modeling for signal analysis: Raw- waveform CNNs are developed for identification of gender, and for sub challenges of ComParE2018 (Infant vocalization, heart beat and Self assessed affect).

• Raw-waveform CNNs are developed for end-to-end signal recovery tasks in spike estimation and source separation.

Indian Institute of Technology Madras Chennai, India

Research Assistant in Speech and Music Technology Lab 2013 – December 2018

• Phase-based methods for signal analysis.

• Analysis of High Resolution Property of Group Delay function: Developed a theoretical basis forthe high spectral resolution of group delay.

• Pitch estimation from speech using Modified Group Delay-gram: Developed an algorithm forpitch estimation through application of GCT on reinforced modified group-delay-gram.

• -CompMusic Project (Collaborated with five universities (ERC Grant)).

  • Group Delay Based Music Source Separation Using Deep Recurrent Neural Networks: Developed a source separation-driven algorithm for onset detection in musical mixtures consists of vocal, violin and Mridangam.

  • An algorithm combining envelope detection using Hilbert transform and minimum phase group delay function is developed for percussion solos.

University of Southern California, United States

Visiting Researcher at Signal Analysis and Interpretation Lab March 2017 – May 2017

Joint segmentation and clustering for dyadic speaker diarization using IB system and denoising of vocal interactions.

Massachusetts Institute of Technology, United States

Visiting Researcher at Sur Lab March 2017

Computational Brain Research Project: Developed a phase-based signal processing algorithm for spike estimation exploiting the high resolution nature of group delay function.