Skip to content
GitHub

Quickstart: lexigram-multimedia-tts

Synthesize speech from text, wired through the Lexigram DI container. The default backend hits a local HTTP reference server, so it works with zero API keys — no cloud account, no secret setup.

Install

Section titled “Install”
Terminal window
uv add lexigram-multimedia-tts

Optional extras pull in a backend’s SDK or server runtime:

Terminal window
uv add "lexigram-multimedia-tts[elevenlabs]"        # ElevenLabs API
uv add "lexigram-multimedia-tts[openai]"            # OpenAI TTS API
uv add "lexigram-multimedia-tts[chatterbox-server]" # local Chatterbox (torch)
uv add "lexigram-multimedia-tts[kokoro-server]"     # local Kokoro-82M
uv add "lexigram-multimedia-tts[f5-tts-server]"     # local F5-TTS voice cloning (torch)
uv add "lexigram-multimedia-tts[piper-server]"      # local Piper (ONNX, CPU)

lexigram-multimedia-tts depends on lexigram and lexigram-contracts (installed automatically). aiohttp is a hard dependency used by every backend.

Minimal Working Example

Section titled “Minimal Working Example”
import asyncio


from lexigram import Application
from lexigram.contracts.multimedia import TTSProvider, TTSRequest
from lexigram.di.module import Module, module
from lexigram.multimedia.tts import AudioTTSModule




@module(imports=[AudioTTSModule.configure()])
class AppModule(Module):
    pass




async def main() -> None:
    async with Application.boot(modules=[AppModule]) as app:
        tts: TTSProvider = await app.container.resolve(TTSProvider)
        result = await tts.generate(TTSRequest(text="Hello from Lexigram", voice="alloy"))
        if result.is_ok():
            asset = result.unwrap()  # MediaAsset — raw audio bytes
            print(asset.mime_type, asset.has_bytes)




asyncio.run(main())

Point the local-http backend at any TTS HTTP server exposing POST /generate (the simplest path is a Coqui/Piper-style server on port 5002).

What Just Happened

Section titled “What Just Happened”
  • AudioTTSModule.configure() created a DynamicModule carrying AudioTTSProvider, exporting the TTSProvider contract and TTSGenerationTask.
  • Application.boot() ran the provider lifecycle:
    • register — TTSConfig is bound as a singleton; the provider builds the configured backend (LocalHttpTTSProvider by default) and binds it as TTSProvider, plus a TTSGenerationTask wrapper. API backends additionally resolve their API key from the secrets backend by name.
    • boot — no extra async I/O (all wiring happened in register()).
  • tts.generate(TTSRequest(...)) POSTs to {backend_base_url}/generate and returns a Result[MediaAsset, TTSError] — domain failures are values, not exceptions. MediaAsset carries mime_type, provider, bytes_data.

Next Steps

Section titled “Next Steps”