Text to Speech Applied Research

Text to Speech Applied Research

Text to Speech Data Ops

• Collaborated with Text-to-Speech (TTS) Research and Production teams to push experimental voices to production.• Built experimental Wavenet and LSTM voices for various languages.• Trained and compared the performance of experimental TTS prosody models for diverse languages.• Evaluated voice-performance and migrated high-scoring voices to production pipeline.• Analyzed and implemented changes to improve script selection tool.• Headed several long-term research projects, consistently producing high-quality deliverables.• Utilized Agile and Scrum methodologies to manage and optimize workflow.• Tested, documented, and made recommendations for integration of research tools into TTS Pipeline.• Created recording scripts and collaborated with external vendors to process new speaker data.• Regularly used and debugged the end-to-end TTS pipeline and related tools.

• Built Average Voice Models (AVMs) using Deep Neural Networks (DNNs), and tuned model hyperparameters.• Experimented with various DNN types including: feedforward, LSTM and BLSTM.• Created speaker-specific synthetic voices via voice conversion using sprocket, speaker adaptation using LHUC, and speaker embeddings (i-vectors/speaker codes).

• Annotated textual data for grammatical, usage, spelling and mechanical errors.• Analyzed data to improve functionality of annotation tool.

• Classified and annotated linguistic data.• Transcribed and labeled text for disambiguation, expansion, and text normalization.

• Provided undergraduate and graduate ESL learners with conversation practice, grammar lessons, and editing services.

• Researched evidence of Morpho-Syntactic change in English-Bengali code-switching.• Performed literature review, data entry, data organization, and data analysis.• Developed and presented an academic poster summarizing preliminary findings at the Summer.• Research Opportunities Program (SROP) Research Symposium.

• Performed data analysis, validation processes, and categorization of verb phrase ellipsis.• Organized data into syntactic trees, using Tregex.