[Enhancement] Adding support for Openai TTS models

[amitsnow]

Summary

Adds Text-to-Speech (TTS) support for OpenAI and Azure OpenAI models in SyGra. Generate high-quality audio from text with voices, audio formats, and variable speed control. Base64-encoded audio is automatically saved to organized file structures.

Explain the features implemented:

• TTS Models: TTS model support for both OpenAI and Azure OpenAI
• Auto File Saving: Base64 audio data URLs automatically saved to task_dir/output_dir/audio/
• Smart Processing: Recursively handles nested structures, replaces data URLs with file paths
• Multi-format Support: 9 audio formats + 8 image formats

Performance Impact

• Minimal memory impact (base64 converted to files during checkpoint writing)

How to Test

Unit Tests

# Run TTS tests
pytest tests/core/models/test_openai_client.py::test_create_speech_async -v
pytest tests/utils/test_audio_utils_save.py -v
pytest tests/utils/test_multimodal_processor.py -v

End-to-End Test

1. Add the model config in models.yaml:

      tts_openai:
      model: tts
      output_type: audio  # This triggers TTS functionality
      model_type: azure_openai  # Use azure_openai or openai model type
      api_version: 2025-03-01-preview
      parameters:
        voice: "alloy"
        response_format: "wav"

2. Add URL and API_KEY:
   URL and api_key should be defined at .env file as SYGRA_TTS_OPENAI_URL and SYGRA_TTS_OPENAI_TOKEN

3. Run pipeline

4. File Structure Output

task_dir/
├── output/audio/
│   ├── record_001_generated_audio_0.mp3
│   └── record_002_generated_audio_0.mp3
└── output.json

Checklist

• Lint fixes and unit testing done
• End to end task testing
• Documentation updated

Notes

• Text limit: 4096 characters per request (OpenAI constraint)

[psriramsnc]

LGTM 🚀