Staff Engineer

Audience Communications (A Knowles Company)

I was part of the VoiceIQ Algorithms Productization team.

VoiceIQ was Audience’s solution for low power spoken keyword recognition (Voice Wake) in smartphones and home devices.

My role included the following activities:

• Data pre-processing and fine-tuning of Machine Learning based algorithms.
• Evaluating performance in various distractor scenarios (car, babble, rain etc.)
• On device training algorithm improvements based on customer requirements.
• Generic algorithm improvements/research to improve performance.
• Single point of contact for VoiceIQ algorithms team in India.

Few highlights of my work:

• Two design wins from major mobile OEMs in 2017-18, with in-lab accuracy for voice wake improved upto 94% in babble distractor scenarios.
• One US Patent resulted from my work on variable thresholding based on SNR measurements at run-time.
• My research work included modelling multi-microphone topologies for IoT devices. With this I worked on predicting microphone beamforming performance in different cases for diffuse and directional distractors.

Languages/Tools/Frameworks: Python, C, C++, Adobe Audition, Audacity, HTKToolkit, Visual Studio Code, Git.
Theory: Speech, Voice wake, Signal Processing, Beamforming.