Rime SDK for Python

Requires Python 3.11 or later. This is an alpha release. Install it from PyPI with uv:

uv add "rimelabs-sdk==0.1.0a1"

import asyncio
from rimelabs_sdk import Rime, AudioFormat

async def main():
    async with Rime(api_key="your-api-key") as client:
        async with client.tts.stream(
            "Hello. This is Rime.",
            voice="clementine",
            audio_format=AudioFormat.PCM_24000,
        ) as audio:
            with open("speech.pcm", "wb") as output:
                async for chunk in audio:
                    output.write(chunk)
            print(audio.request_id)

asyncio.run(main())

Omit api_key to read RIME_API_KEY. The default model is coda. The SDK selects its standard endpoint. Only coda is currently supported. For a custom deployment of that model, set endpoint:

client = Rime(model="coda", endpoint="coda.api.customer-name.rime.ai")

Use a hostname with an optional port, such as host:8443. Omit the scheme and path. Connections always use TLS; the default port is 443. The endpoint applies to speech, voices, and languages for this client.

The default voice is clementine; the default language is en. tts.stream() also accepts AsyncIterable[str]. Do not await the factory. It starts work when you enter its async context or request the first chunk. Use the stream context to release resources when you exit a loop early.

await client.voices.list(language="en") returns voice names. await client.languages.list() returns language codes. Both calls accept a keyword timeout in seconds.

Rime(timeout=30) sets an overall timeout for operations. An operation can replace it with timeout=10 or disable it with timeout=None. Omitting the operation option inherits the client setting. Internal connection and progress limits still apply. An overall timeout continues during caller pauses.

audio.format is a read-only AudioFormat with encoding, sample_rate, and channels. PCM_24000 is raw signed 16-bit little-endian mono PCM at 24 kHz. MULAW_8000 is raw mono G.711 mu-law at 8 kHz. Neither includes a file header. Chunks preserve complete sample frames. Chunk sizes are not fixed.

await audio.cancel() cancels one operation. await client.close() cancels all client work. Both are idempotent. A client belongs to one process and event loop. Normal iterator completion means that the final service status was successful. Partial audio can precede a typed error. Synthesis is never replayed.

Catch RimeError or one of RimeAuthenticationError, RimePermissionError, RimeInputError, RimeResourceLimitError, RimeUnavailableError, RimeTimeoutError, RimeAudioFormatError, RimeCancelledError, or RimeStreamError. Errors expose a message and optional request_id. Cancellation of the caller's Python task keeps asyncio.CancelledError.

The SDK sends the API key as a bearer token over TLS.

Development for contributors

These instructions require access to the source repository. Run the commands from its python/ directory:

uv sync --locked --dev
uv run ruff check src tests
uv run ruff format --check src tests
uv run mypy src
uv run pytest
uv build

License

The SDK is licensed under the MIT License.