Add support for transcript only scribes

[shivankacker]

This PR introduces significant changes to environment variables.

Removed

• SCRIBE_API_PROVIDER
  • This variable has been removed entirely.
  • Providers must now be specified directly before each model name.

Renamed

• SCRIBE_AUDIO_MODEL_NAME → SCRIBE_TRANSCRIBE_MODEL_NAME
  • This variable now expects the provider to be prefixed to the model name.
  • Example:

    SCRIBE_TRANSCRIBE_MODEL_NAME=google/gemini-2.5-flash

Updated Behavior

• SCRIBE_CHAT_MODEL_NAME
  • This variable now also expects the provider to be prefixed to the model name.

Added

• SCRIBE_TRANSCRIBE_LANGUAGE
  • This variable dictates the final transcript language, if google models are used to transcribe. Not assigning this or keeping blank would mean the transcript is in the same language as the source audio.
  • I recommend setting this to en-US as our requirement is to translate all langs to english.

Supported Transcription Models

OpenAI

Examples:

openai/whisper-1
openai/gpt-4o-transcribe

Google

• Google transcription uses standard Gemini LLMs
• Examples :

google/gemini-2.5-flash