speech insight

/overview

a complete end-to-end speech recognition (asr) system that segments long audio files (like meetings) and transcribes them.

/implementation

uses voice activity detection (vad) to smartly segment audio into sentence-level clips. the custom asr model is a pytorch-based crnn (cnn + gru + ctc) trained on librispeech. users interact with an intuitive dashboard built with streamlit.

/the tech stack