Investigations

AI Audio Suite

Python June 2025

Production-ready Docker deployment of MeloTTS and Whisper with a Go API gateway, React frontend, and GPU acceleration.

This combines two AI audio capabilities into one deployable system: MeloTTS for text-to-speech and OpenAI's Whisper for speech-to-text. A Go API gateway sits in front of both services, and a React frontend provides the user interface.

The Docker setup handles the complexity — there are separate compose configs for CPU, GPU, AWS, and English-only deployments. The smart start script auto-detects NVIDIA GPUs and configures accordingly. With GPU acceleration, a 1-hour lecture transcribes in about 3 minutes versus 30 on CPU.

MeloTTS supports multiple English accents (US, British, Indian, Australian) with configurable speed. Whisper handles models from tiny (39M parameters) to large (1.5B). The Go gateway provides health monitoring, Swagger documentation, and WebSocket streaming for real-time transcription.

For AWS, there are spot instance scripts that spin up GPU machines on demand — processing that 1-hour lecture costs about $0.006. The whole architecture is designed for cost-effective batch processing.

What this project demonstrates

MeloTTS + Whisper in production-ready Docker deployment
Go API gateway with health monitoring and Swagger docs
Auto-detect GPU: 10-20x speedup with NVIDIA acceleration
Multiple accents and Whisper model sizes (39M to 1.5B params)
AWS spot instance processing at ~$0.006 per hour of audio
WebSocket streaming for real-time transcription

Technologies

AIDockerTTSWhisperGoGPUAWS

© 2026 Jonathan Leahy · v0.8.1-31-g196fa14