It's hard to believe the KaggleX mentorship program is coming to an end. Over the past several weeks, I've had the pleasure of mentoring Sakshi as she took on the challenge of building an automatic speech recognition (ASR) system.
When we first started, Sakshi had a basic understanding of machine learning but lacked hands-on experience with ASR. Together, we outlined an ambitious project to build an end-to-end ASR pipeline using deep learning.
There were plenty of hurdles along the way. We spent many late nights debugging errors in the data preprocessing (loading MP3 in Kaggle) and model training code. Improving the performance metric was a tedious process of trial and error. But Sakshi never became discouraged. She routinely impressed me with her persistence and ability to quickly pick up new technical skills (Initially, she was lazy, but with time, she became super focused. )
In the end, we succeeded in building 2 ASR models and web applications that exceeded our initial expectations. The model performs well on noisy audio and recognizes a variety of speakers with high accuracy. The most important part was documenting the code and analysis. I think Sakshi is working on writing a Medium blog so that everyone can understand what went through in detail.
Participating in this mentorship program has been incredibly rewarding for both of us. I'm grateful to have had the opportunity to mentor such a dedicated mentee. Sakshi is now well-equipped to continue pursuing her interests in speech and language processing. I have no doubt she has a bright future ahead.
DagsHub: https://dagshub.com/SakshiRathi77/hindiSpeechPro-Automatic-Speech-Recognization
Presentation: https://www.loom.com/share/ba384456d0ad4db5b2a94427ab6b6441?sid=fc752498-d907-4f87-a9e3-20920a021c63