| 1234567891011121314151617181920212223242526272829 |
- name: stt-vocab
- repo: https://github.com/pierrenode/hermes-stt-vocab
- sha: 54e83f9289315af6d47c7c6fdeb5714a08b35262
- subdir: stt-vocab
- description: >-
- Puts the names a speech-to-text model mis-hears into the transcription prompt Hermes sends with
- each voice message: terms you list (globally or per surface) and names found in this profile's
- memories/USER.md and MEMORY.md, most important last, within the 896 characters Hermes keeps for
- Whisper-family backends; your stt.prompt stays in front. Notes written in German, Spanish,
- French, Italian, Dutch, Portuguese or Turkish can be named in the languages setting.
- Disclosure — register() adds one pre_transcription hook that returns a new prompt string;
- memory-derived names go only to local backends (faster-whisper, local_command) unless
- allow_remote is set, and the memory files are not read otherwise; reads at most 256 KB per
- memory file; no tools, network access, subprocesses or writes; Hermes uploads the prompt with
- the audio to the configured STT provider.
- maintainer: pierrenode
- tier: community
- category: voice
- requires_hermes: ">=0.21"
- docs_url: https://github.com/pierrenode/hermes-stt-vocab
- version: "1.1.0"
- image: https://raw.githubusercontent.com/pierrenode/hermes-stt-vocab/6145d065d50f196d51b3ab80fe4a76fb2827c115/docs/card.png
- platforms: []
- capabilities:
- provides_tools: []
- provides_hooks:
- - pre_transcription
- provides_middleware: []
- requires_env: []
|