Why

Hermes already had a flexible multi-provider speech-to-text pipeline, but it did not yet support SenseAudio as an STT backend.

This branch adds SenseAudio as an additional provider within the existing transcription_tools.py dispatcher so it can participate in the same provider selection, config, and fallback flow as the current local, Groq, OpenAI, and Mistral implementations.

Summary

1) Add SenseAudio as a first-class STT provider

• Extends tools/transcription_tools.py with a new senseaudio provider path
• Adds SenseAudio-specific defaults and constants:
  • DEFAULT_SENSEAUDIO_STT_MODEL
  • SENSEAUDIO_BASE_URL
  • SENSEAUDIO_MODELS
• Uses the current SenseAudio STT API through the existing OpenAI-compatible client path

2) Wire SenseAudio into provider selection

• Updates _get_provider() so an explicitly configured stt.provider: senseaudio is respected
• Adds validation for SenseAudio availability:
  • openai package must be installed
  • SENSEAUDIO_API_KEY must be set
• Extends auto-detection so SenseAudio can be selected when higher-priority STT providers are unavailable

3) Add provider-specific transcription implementation

• Introduces _transcribe_senseaudio(file_path, model_name)
• Loads provider-specific config from the existing STT config structure
• Creates a provider-scoped OpenAI-compatible client with SenseAudio base URL and API key
• Returns the same normalized result shape used by the other STT providers

4) Keep the existing STT architecture and UX consistent

• Integrates SenseAudio into transcribe_audio() using the same dispatcher pattern as the other providers
• Auto-corrects unsupported model names to the default SenseAudio STT model
• Updates the “no provider available” error message to include SenseAudio as another supported option

Safety and Regression Notes

• No new tool surface was introduced; this change extends the existing STT tool path only.
• The implementation stays within the current single-file multi-provider STT architecture.
• SenseAudio is added as another provider option rather than changing the behavior of existing providers.
• Regression risk is limited because the dispatcher structure and return contract remain unchanged.

Files Changed

Updated:

• tools/transcription_tools.py

Test Evidence

Automated tests

Ran the existing STT tool test suite against the branch:

• tests/tools/test_transcription_tools.py

Result: 71 passed in 13.38s

Manual smoke tests