Ramya Viswanathan Email & Phone Number

Bengaluru, Ka, In

Bangalore Urban, Karnataka, India

Leading three different modules:Deployment:1) Keyword Spotting in Samsung devices (Galaxy Mobile, Galaxy Watch, Refrigerator, etc.,)2) Voice Activity Detection in resource constrained devices.Research:1) Audio-LLM and Audio Foundation Model Development for different Audio Tasks (TTS, Source Separation).

Bangalore Urban District, Karnataka, India

Wakeup system development and deployment in Visual display and Digital Appliances (Samsung TVs, FamilyHub, Vaccum Cleaner, etc.,)

May 2018 ~ Jul 2018WFST Search: Function API design & definitions for implementation of WFST search in eASR engine.June 2018 ~ May 2019Keyword Spotting Engine Development : Implementation of KWS engine for Smart Speakers. Different AM architectures & training techniques (sMBR, Adaptation) employed. Data Argumentation using RIR simulators for far-field & low SNR scenarios. Decoder & Confidence measure alterations.June 2019 ~ Present:Similar Word Pattern Mining for Keyword Spotting Task

Jun 2016 ~ Dec 2016Tizen eASR development: Embedded ASR (eASR) development for Tizen devices (Tizen Z2, Gear Watch, TV). Decoder optimizations & Confidence score boosting with dynamic pronunciation addition.Jan 2017 ~ Apr 2017ASR Developement: Decoder parameters optimization for trade off between accuracy and latency.May 2017 ~ Dec 2017Accent Determination: System to determine the accent of the user using Linguistic clues from set of sentences/words and contextual clues.Jan 2018 ~ May 2018Wakeup Development: Stage - 2 AP On-device Wakeup eASR engine development for keyword verification. Confidence Measure & decoder enhancements specific to Keyword phonemes. DNN architecture changes for trade off between CPU load and Equal Error Rate (EER).

Jun 2014 ~ Dec 2014:Speech Recognition: Gaussian Mixture Models (GMM) Trainings using CMU Sphinx Toolkit - Sphinxtrain. Understanding of Decoder, Language Models, CMU embedded engine toolkits (Pocketsphinx, Sphinxbase).Jan 2015 ~ Jun 2015Confidence Measure (CM) & Embedded ASR (eASR) development: Embedded speech recognition engine development with support of Confidence measure to reject/accept the output of the system. Exploration & implementation of different CM techniques: Classifier (SVM), Posterior Probability, Utterance Verification (UV).Jul 2015 ~ Dec 2015Neural Networks: DNNs, LSTM training using Kaldi toolkit. Confidence Measure Enhancements: Speech/Non-speech Identification, Posterior Phoneme sequence verification.Jan 2016 ~ May 2016eASR Development: Implementation & Porting of Kaldi formats in Sphinx formats. Latency reduction in decoder search tree for Kaldi implementation.

Postgraduate Degree, Artificial Intelligence, Excel

Class Topper in the batch.

Bachelor Of Engineering (B.E.), Computer Science, Gpa: 8.78/10

Activities and Societies: 1) Volunteer and coordinator of events in College Festivals (KURUKSHETRA'11, ABACUS'12) 2) An Active member.