Audio Transcriber: Fully Offline Speech to Text
Published:
Timeline: June 2026 - July 2026
Project Description: Drop an audio file into your browser and get back a transcript, with an optional English translation, processed entirely on your own machine. No audio leaves the computer and no API keys are needed. It was built for Bengali call recordings that mix in English, but works for any language Whisper supports. A standard mode uses whisper.cpp with the large-v3 model, and a high-quality mode adds faster-whisper and pyannote speaker diarization for speaker-labelled transcripts. Built with Python and Flask.
Project Github Link: audio-transcriber