dave-parsons
commented
4 months ago Oh - and forgot - Multi-language is key for me. Will send the raw data for transcription in Spanish and English minimum.
Open dave-parsons opened 4 months ago
dave-parsons
commented
4 months ago Oh - and forgot - Multi-language is key for me. Will send the raw data for transcription in Spanish and English minimum.
Potential feature request:
Every day, I'm on 8 hours of calls talking about stuff. Teams already allows meeting transcriptions. Got me thinking about something far more universal for Home Assistant.
1) Goal: "Noter" application invokes transcription of continuous play audio (in chunks up to 1 hour) that are text transcribed. Resulting data set is "summarized" by LLM and made available in categorized repository. 2) Requirement: Inspired and excited by Wyoming Sat Wake Word ByPass and the Whisper-Live. a) Trigger a session of any length started and stopped by user (voice, automation, button, etc) b) Transcribe raw STT in human readable categories (e.g. by time/date in repository that can be browsed, reviewed. c) Send raw transcription to LLM (Local future) to summarize and store in human readable categories related to raw (in "b") 3) Idea / Feature request. Utilize whisper addon. and enable a) triggered start/stop transcription service b) store results in HA storage/media repository c) blueprint setup to automate sending raw to LLM pipeline, get summary, and store in repository for use as user desires.
Working on ways to use Wyoming-satellite, whisper, Home Assistant, and LLM to make that real for me now. Any pointers appreciated!