If you've built anything with text-to-speech lately, you know the pain: every provider has its own SDK, its own authentication pattern, its own audio format quirks. Switch from OpenAI's TTS to ElevenLabs and you're rewriting integration code. SpeechSDK, a new open-source project from Jellypod, Inc., tries to fix that with a single interface that wraps 12 different voice AI providers.
The SDK supports 25+ models across OpenAI (including gpt-4o-mini-tts), ElevenLabs, Deepgram, Cartesia, Google Gemini, Hume, Fish Audio, Murf, Mistral, Resemble, Unreal Speech, and fal. You install one npm package (@speech-sdk/core), point it at whichever provider you want, and get audio back in a consistent format. Switching providers is a one-line change to the model string.
How It Actually Works
You bring your own API keys for whatever providers you use. A typical call looks like:
const result = await generateSpeech({
model: 'openai/gpt-4o-mini-tts',
text: 'Hello from SpeechSDK!',
voice: 'alloy',
});
Swap openai/gpt-4o-mini-tts for elevenlabs/eleven_multilingual_v2 and the rest stays the same. The SDK handles format differences behind the scenes - OpenAI returns MP3, Cartesia returns WAV, Google returns WAV - and exposes a unified response with lazy base64 conversion (meaning it only computes the base64 string if you actually access it, saving memory).
Other practical details: zero runtime dependencies beyond a retry library, built-in exponential backoff for failed requests, AbortSignal support for cancellation, and voice cloning for providers that support it. It runs in Node.js, edge runtimes, and browsers.
Who This Is For
This is a developer tool, not a consumer product. It's most useful if you're building a product that uses TTS and want the flexibility to switch providers without rewriting code - say, to chase better pricing, lower latency, or a voice that fits your use case better. The MIT license means no strings attached.
Jellypod also offers a separate "Speech Gateway" product that adds queuing, analytics, and quality processing on top, but the core SDK is fully standalone and free.