A high-performance Text-to-Speech (TTS) service built with FastAPI and Coqui TTS.

VoxRaga delivers natural-sounding speech synthesis with support for multiple languages and voices. Built for running locally and on Kubernetes clusters, it offers a RESTful API that integrates seamlessly with existing speech processing pipelines.

• High-quality speech synthesis using state-of-the-art neural models
• Multi-language and multi-voice support
• Adjustable speech parameters (speed, pitch, format)
• REST API with JSON interface
• GPU-accelerated inference
• Kubernetes-ready containerization
• Prometheus metrics and health monitoring

• Python 3.11+
• Kubernetes cluster with GPU nodes (for cloud deployment)
• Docker or Podman
• espeak or espeak-ng (for phonemization)

# Clone the repository
git clone https://github.com/yourusername/vox-raga.git
cd vox-raga

# Setup development environment
make setup-dev

# Run development server
make dev

# Build Docker image
make build

# Run locally
make run

# Login to ACR
make acr-login

# Build and push in one step
make acr-push

VoxRaga is deployed as a KServe InferenceService, which provides scaling, monitoring, and routing capabilities.

# Apply Kubernetes manifests
kubectl apply -f k8s/inferenceservice.yaml

# Check deployment status
kubectl get inferenceservices

The deployment creates a KServe InferenceService that automatically scales based on demand and provides a RESTful endpoint for clients to consume.

VoxRaga is configured through environment variables:

curl -X POST http://localhost:8888/synthesize \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hello, this is a test of the text to speech system.",
    "options": {
      "language": "en",
      "voice": "p225",
      "speed": 1.0,
      "format": "wav"
    }
  }' --output test.wav

curl -X GET http://localhost:8888/voices

curl -X GET http://localhost:8888/languages

VoxRaga includes comprehensive test suites:

# Run all tests
make test

# Try sample client
cd samples
python test_tts.py --list-voices
python test_tts.py --voice p225 --format wav

For optimal performance:

• Enable hardware acceleration where available
• Set MODEL_COMPUTE_TYPE=float16 for faster inference
• Consider models with lower latency for real-time applications

VoxRaga exposes Prometheus metrics at the /metrics endpoint for monitoring:

• Request latency and throughput
• Model inference time
• Cache hit/miss rates
• Resource utilization

This project is licensed under the MIT License - see the LICENSE file for details.