feat(stt): pass local initial_prompt to faster-whisper by BlackishGreen33 · Pull Request #50684 · NousResearch/hermes-agent

BlackishGreen33 · 2026-06-22T07:46:53Z

What does this PR do?

Adds stt.local.initial_prompt support for the local faster-whisper STT provider.

This lets users bias local transcription output, for example asking Chinese audio to be written in Simplified Chinese. The prompt is only passed when it is configured as a non-empty string, so the current default behavior stays unchanged.

Related Issue

Fixes #50181

Changes Made

Read initial_prompt from the existing stt.local config block.
Pass initial_prompt to faster-whisper only when it is a non-empty string.
Keep the existing language, beam size, and CUDA fallback behavior unchanged.
Added tests for configured and blank initial_prompt values.
Added the new config key to cli-config.yaml.example.

Configuration Example

stt:
  local:
    language: zh
    initial_prompt: "以下是普通话的句子，使用简体中文。"

Verification

$ ../hermes-agent/venv/bin/python -m pytest tests/tools/test_transcription.py -q
............................                                             [100%]
28 passed in 0.66s

$ git diff --check
# no output

Checklist

I searched for existing PRs to make sure this is not a duplicate.
This PR only changes local faster-whisper STT initial_prompt handling.
I added focused tests for the new config behavior.
I updated cli-config.yaml.example for the new config key.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

teknium1

Thanks for the focused implementation. The forwarding behavior is still missing on current main: tools/transcription_tools.py:1131-1142 builds the faster-whisper kwargs with language and beam size only.

Problems

The PR documents stt.local.initial_prompt only in cli-config.yaml.example; it does not add the key to the runtime configuration schema at hermes_cli/config.py:2132-2135. New config options should be represented in DEFAULT_CONFIG, so config introspection and default merging know the setting exists.

Suggested changes

Add "initial_prompt": "" under DEFAULT_CONFIG["stt"]["local"] and cover the real config-resolution path in addition to the mocked forwarding tests.

This is an automated hermes-sweeper review.

BlackishGreen33 · 2026-07-18T12:29:41Z

Follow-up test isolation fix in 142034a0b: test_accepts_null_local_config now removes HERMES_LOCAL_STT_LANGUAGE before asserting the exact faster-whisper kwargs. I reproduced the issue first with the env set to fr (language: fr leaked into the call), then verified the isolated test and the focused transcription/config suites pass.

@zombopanda

Class-level fix for the 'STT transcribes the wrong language' issue family (#55551, #50181 and siblings). Previously language handling was per-provider chaos: local honoured stt.local.language, Groq/OpenAI/Mistral/DeepInfra sent no language hint at all, xAI silently forced 'en', ElevenLabs used its own language_code key, and there was no global setting. - New _resolve_stt_language() helper: stt.<provider>.language > stt.language (new global key) > HERMES_LOCAL_STT_LANGUAGE > auto-detect. - Threaded through ALL providers: local, local_command, groq, openai, mistral, xai, elevenlabs, deepinfra (shared OpenAI handler), command providers, and plugin dispatch. - xAI no longer forces English when nothing is configured (auto-detect). - Mistral Voxtral now receives a language hint when configured. - stt.groq.model is now honoured from config (previously env-only). - DEFAULT_CONFIG gains stt.language, stt.groq, stt.xai, stt.mistral.language. - Tests: tests/tools/test_stt_language_resolution.py (11 tests, sabotage- verified) + full transcription suite green (236 passed). Builds on cherry-picked contributor work from #19786 (@zombopanda), #23161 (@materemias), #50684 (@BlackishGreen33).

@zombopanda

Class-level fix for the 'STT transcribes the wrong language' issue family (#55551, #50181 and siblings). Previously language handling was per-provider chaos: local honoured stt.local.language, Groq/OpenAI/Mistral/DeepInfra sent no language hint at all, xAI silently forced 'en', ElevenLabs used its own language_code key, and there was no global setting. - New _resolve_stt_language() helper: stt.<provider>.language > stt.language (new global key) > HERMES_LOCAL_STT_LANGUAGE > auto-detect. - Threaded through ALL providers: local, local_command, groq, openai, mistral, xai, elevenlabs, deepinfra (shared OpenAI handler), command providers, and plugin dispatch. - xAI no longer forces English when nothing is configured (auto-detect). - Mistral Voxtral now receives a language hint when configured. - stt.groq.model is now honoured from config (previously env-only). - DEFAULT_CONFIG gains stt.language, stt.groq, stt.xai, stt.mistral.language. - Tests: tests/tools/test_stt_language_resolution.py (11 tests, sabotage- verified) + full transcription suite green (236 passed). Builds on cherry-picked contributor work from #19786 (@zombopanda), #23161 (@materemias), #50684 (@BlackishGreen33).

teknium1 · 2026-07-28T03:30:13Z

Merged via PR #73067 — your commit was cherry-picked onto current main with your authorship preserved in git history. Thanks for the contribution! The fix is now part of a unified language resolver covering every STT provider.

@zombopanda

Class-level fix for the 'STT transcribes the wrong language' issue family (NousResearch#55551, NousResearch#50181 and siblings). Previously language handling was per-provider chaos: local honoured stt.local.language, Groq/OpenAI/Mistral/DeepInfra sent no language hint at all, xAI silently forced 'en', ElevenLabs used its own language_code key, and there was no global setting. - New _resolve_stt_language() helper: stt.<provider>.language > stt.language (new global key) > HERMES_LOCAL_STT_LANGUAGE > auto-detect. - Threaded through ALL providers: local, local_command, groq, openai, mistral, xai, elevenlabs, deepinfra (shared OpenAI handler), command providers, and plugin dispatch. - xAI no longer forces English when nothing is configured (auto-detect). - Mistral Voxtral now receives a language hint when configured. - stt.groq.model is now honoured from config (previously env-only). - DEFAULT_CONFIG gains stt.language, stt.groq, stt.xai, stt.mistral.language. - Tests: tests/tools/test_stt_language_resolution.py (11 tests, sabotage- verified) + full transcription suite green (236 passed). Builds on cherry-picked contributor work from NousResearch#19786 (@zombopanda), NousResearch#23161 (@materemias), NousResearch#50684 (@BlackishGreen33).

@zombopanda

Class-level fix for the 'STT transcribes the wrong language' issue family (NousResearch#55551, NousResearch#50181 and siblings). Previously language handling was per-provider chaos: local honoured stt.local.language, Groq/OpenAI/Mistral/DeepInfra sent no language hint at all, xAI silently forced 'en', ElevenLabs used its own language_code key, and there was no global setting. - New _resolve_stt_language() helper: stt.<provider>.language > stt.language (new global key) > HERMES_LOCAL_STT_LANGUAGE > auto-detect. - Threaded through ALL providers: local, local_command, groq, openai, mistral, xai, elevenlabs, deepinfra (shared OpenAI handler), command providers, and plugin dispatch. - xAI no longer forces English when nothing is configured (auto-detect). - Mistral Voxtral now receives a language hint when configured. - stt.groq.model is now honoured from config (previously env-only). - DEFAULT_CONFIG gains stt.language, stt.groq, stt.xai, stt.mistral.language. - Tests: tests/tools/test_stt_language_resolution.py (11 tests, sabotage- verified) + full transcription suite green (236 passed). Builds on cherry-picked contributor work from NousResearch#19786 (@zombopanda), NousResearch#23161 (@materemias), NousResearch#50684 (@BlackishGreen33).

@zombopanda

Class-level fix for the 'STT transcribes the wrong language' issue family (NousResearch#55551, NousResearch#50181 and siblings). Previously language handling was per-provider chaos: local honoured stt.local.language, Groq/OpenAI/Mistral/DeepInfra sent no language hint at all, xAI silently forced 'en', ElevenLabs used its own language_code key, and there was no global setting. - New _resolve_stt_language() helper: stt.<provider>.language > stt.language (new global key) > HERMES_LOCAL_STT_LANGUAGE > auto-detect. - Threaded through ALL providers: local, local_command, groq, openai, mistral, xai, elevenlabs, deepinfra (shared OpenAI handler), command providers, and plugin dispatch. - xAI no longer forces English when nothing is configured (auto-detect). - Mistral Voxtral now receives a language hint when configured. - stt.groq.model is now honoured from config (previously env-only). - DEFAULT_CONFIG gains stt.language, stt.groq, stt.xai, stt.mistral.language. - Tests: tests/tools/test_stt_language_resolution.py (11 tests, sabotage- verified) + full transcription suite green (236 passed). Builds on cherry-picked contributor work from NousResearch#19786 (@zombopanda), NousResearch#23161 (@materemias), NousResearch#50684 (@BlackishGreen33).

feat(stt): pass local initial_prompt to faster-whisper

a5fbe95

Copilot AI review requested due to automatic review settings June 22, 2026 07:46

Copilot AI reviewed Jun 22, 2026

alt-glitch added type/feature New feature or request tool/tts Text-to-speech and transcription P3 Low — cosmetic, nice to have labels Jun 22, 2026

alt-glitch mentioned this pull request Jun 28, 2026

fix(stt): wire stt.local.initial_prompt through to faster-whisper #54202

Closed

3 tasks

🔧 chore(stt): merge upstream main

cb1c332

teknium1 reviewed Jul 14, 2026

View reviewed changes

Comment thread cli-config.yaml.example

teknium1 added sweeper:risk-compatibility Sweeper risk: may break existing users, config, migrations, defaults, or upgrades sweeper:blast-moderate Sweeper blast radius: moderate — a subsystem or single platform labels Jul 15, 2026

BlackishGreen33 added 2 commits July 18, 2026 20:18

🐛 fix(stt): declare local initial prompt default

55b5594

✅ test(stt): isolate null config from language env

142034a

GottZ mentioned this pull request Jul 26, 2026

STT: 支持 initial_prompt 配置以偏置简体/繁体中文输出 #50181

Closed

alt-glitch mentioned this pull request Jul 26, 2026

feat(stt): honour initial_prompt and auto-detect language in local transcription #72143

Closed

teknium1 mentioned this pull request Jul 28, 2026

feat(stt): unify language resolution across all STT providers (wrong-language transcription class fix) #73067

Merged

teknium1 closed this Jul 28, 2026

BlackishGreen33 deleted the bg/stt-local-initial-prompt-50181 branch July 28, 2026 04:56

Navigation Menu

Uh oh!

feat(stt): pass local initial_prompt to faster-whisper - #50684

feat(stt): pass local initial_prompt to faster-whisper#50684
BlackishGreen33 wants to merge 4 commits into
NousResearch:mainfrom
BlackishGreen33:bg/stt-local-initial-prompt-50181

BlackishGreen33 commented Jun 22, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

teknium1 left a comment

Uh oh!

Uh oh!

BlackishGreen33 commented Jul 18, 2026

Uh oh!

teknium1 commented Jul 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

BlackishGreen33 commented Jun 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Related Issue

Changes Made

Configuration Example

Verification

Checklist

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

teknium1 left a comment

Choose a reason for hiding this comment

Problems

Suggested changes

Uh oh!

Uh oh!

BlackishGreen33 commented Jul 18, 2026

Uh oh!

teknium1 commented Jul 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

BlackishGreen33 commented Jun 22, 2026 •

edited

Loading