TTS/STT: Feature integration

[nishika26]

Is your feature request related to a problem? Please describe.
Currently, Kaapi only supports text-based user inputs. This limits our ability to:

• Process audio and voice-based inputs from users
• Support voice-enabled chatbot experiences for partners

Describe the solution you'd like:
Phase 1: TTS/STT Exploration & Benchmarking

• Model evaluation: Benchmark existing Indic voice models (Hindi, Tamil, Bengali, etc.) against:

  • Accuracy (WER - Word Error Rate for STT, MOS - Mean Opinion Score for TTS)
  • Latency (real-time vs batch processing)
  • Cost per inference
  • Language/dialect coverage

Phase 2: Platform Integration

• Unified API extension:

  • Add voice model support to /configs endpoint (similar to existing LLM provider configs)
    Extend /llm/call to handle audio inputs/outputs with automatic transcription
    Support both streaming (real-time) and batch audio processing

• Evaluation API enhancement:

  • Add audio-specific evaluation metrics (WER, latency, speaker diarization accuracy)
    Enable quick A/B testing of different TTS/STT providers
    Support voice dataset management (upload, version, annotate)