Local automatic speech recognition (ASR) lets an assistant transcribe meetings on your own hardware — private, and feedable to an LLM for summaries.
The pipeline
audio → ASR engine → time-stamped segments → transcript database → LLM summary / Q&APick an engine by need
| Engine | Strength |
|---|---|
| Whisper | Very accurate; great for after-the-fact transcripts (Whisper with Timestamps) |
| Parakeet | Fast / streaming; good for live capture (Parakeet ASR Setup) |
A common pattern: stream with a fast engine live, then re-process the recording with an accurate one — both writing to the same transcript schema.
Why a shared store
If transcripts land in one database, a meeting captured on one device is instantly available everywhere, and the LLM can answer "what did we decide about X?" across all past meetings.
Pair with a contacts lookup so names in the transcript resolve to people.
Related: Personal AI Assistant Architecture · Whisper with Timestamps · Parakeet ASR Setup