Audio transcription in supervisor route

[Luisotee]

Audio Transcription Support for Supervisor Route

Overview

Add audio transcription capabilities to the supervisor route using Groq's Whisper V3 Turbo API. This centralizes audio processing in the AI API service, eliminating the need for individual transcription handling in client integrations (simulator, WhatsApp, Telegram).

Technical Changes

• Added support for multiple audio formats:
  • Direct support: mp3, mp4, mpeg, mpga, m4a, wav, webm
  • Conversion support: ogg -> mp3
• Implemented audio file handling with temporary storage
• Integrated Groq's Whisper V3 Turbo for transcription
• Added content type detection and validation
• Centralized error handling for audio processing

Dependencies

• Added python-multipart for form data handling
• Added python-ffmpeg for audio conversion
• Added groq for Whisper API access

Configuration

Requires GROQ_API_KEY environment variable

Benefits

• Centralized audio processing
• Consistent transcription quality
• Reduced implementation complexity in clients
• Unified error handling

[luandro]

@Luisotee, it's working great, amazing job! Just don't forget to add the packages you use. For example:

uv add ffmpeg python-multipart

Which will automatically add to the pyproject.toml file and will reflect on every run of the the project. Added on my commit.

A second comment is that the route /api/supervisor/supervisor isn't ideal. Might be a good opportunity to change to /api/classifier or something else.

[Luisotee]

@luandro should be fine now

[luandro]

@Luisotee when testing on the docs page, "send empty value" works when set for message, but for some reason when setting empty value for message an error is throw:

[Luisotee]

@luandro Let me explain the "Send empty value" checkbox:

When checked, it will send an empty argument to the API (like -F 'audio=' or -F 'message='). If you don't want to include either audio or message in the API call, simply leave the checkbox unchecked - this way the parameter won't be included in the request at all.

The error you encountered happened because you were sending both an empty string and no audio file simultaneously, leaving nothing for the API to process.

I've also fixed the dependency issue, so this PR is now ready to be merged.

[luandro]

@Luisotee, thanks for the explanation, but as you can see on the picture I'm not sending an empty message, it has a string: "hello". There is something that the API should process.

[Luisotee]

@luandro The error occurred due to the audio being sent as an empty file rather than being omitted or providing a valid file.

The correct usage is:

• If you don't want to send an audio file, just omit it completely
• If you include the audio parameter, the code expects a valid audio file - otherwise it will (and should) throw this error

About the "Send empty value" checkbox - it's just a default Swagger UI feature. We don't actually use it in our use case, and I haven't found a way to remove it from the interface.