Multi-Provider Text-to-Speech Service

A few teams wanted text-to-speech, and every provider has its own SDK, quirks, and trade-offs in voice quality, latency, and cost. Wiring providers in directly would have spread that mess across every project.

So I built a shared service that gives teams speech through one consistent interface.

The service wraps five providers (OpenAI, Google Cloud, AWS Polly, ElevenLabs, and Deepgram) behind a single API, so switching provider is a config change rather than a rewrite.

It handles both full synthesis, where you generate the whole clip, and low-latency streaming, where playback starts while the audio is still being generated. That covers both batch and real-time use.

Normalise every provider down to one interface so callers never depend on a specific vendor's SDK.

Treat streaming as a first-class mode, not an add-on, so latency-sensitive features feel responsive.

A reusable capability any team can pick up without learning five different APIs, and swap providers freely as their quality, latency, or cost needs change.