Responsibilities include:
- Spearheaded the design and implementation of multilingual neural Text-to-Speech (TTS) architectures for Akan and other African languages, targeting low-resource healthcare and accessibility use cases, including maternal health applications.
- Engineered scalable speech data acquisition pipelines across multiple African countries, incorporating automated quality control, metadata standardization, and integration with cloud storage solutions.
- Developed modular audio preprocessing and segmentation tools using Python (Pydub, NumPy, IPython widgets), enabling rapid annotation and boosting dataset preparation throughput by over 60%.
- Conducted experimental research on prosodic feature modeling abd multi-speaker adaptation in low-resource African language scenarios.
- Built RESTful NLP APIs using FastAPI and integrated evaluation interfaces to deploy TTS/ASR models with real-time inference, speaker evaluation, and performance tracking pipelines.