Speech-to-Text in Real Time

2 points by amohajerani1 2y ago ↗ HN

I want to transcribe meeting notes with following constraints: - Near real time (max 20 seconds delay) - No third party APIs, due to privacy issues.

I looked into whisper, but that starts transcription only after the recording is stopped. I also looked into the browser's native speech recognition, but the browser sends the data to some servers for transcription. Any suggestions?

2 comments

[ 239 ms ] story [ 229 ms ] thread

ajsmitty 2y ago ↗

just feed data into whisper in chunks as the meeting is ongoing? you don't need the whole meeting to be finished in order to process data.

Whisper has to process data in chunks anyways.

dilek 2y ago ↗

the technology you're looking for is "on-device" or "local" speech-to-text

https://cloud.google.com/blog/products/ai-machine-learning/s... https://picovoice.ai/platform/cheetah/ https://github.com/mozilla/DeepSpeech