A new major version release of the Cartesia Python client that overhauls the library structure.
Refer to the [migration guide](https://github.com/cartesia-ai/cartesia-python/discussions/44) for more thorough details on changes.
Features
- Adds support for `model_id=sonic-multilingual` to generate multilingual audio
- Endpoint-specific methods for generating and streaming audio
Breaking changes
- Renames `CartesiaTTS` and `AsyncCartesiaTTS` -> `Cartesia` and `AsyncCartesia`
- Replaces `client.generate` with endpoint-specific methods for Text-to-Speech
- `output_format` must be specified as an `OutputFormat` object, which is a dict specifying the keys: `container`, `encoding` and `sample_rate`
- Both SSE and WebSocket requests no longer return `sampling_rate` in their output. They will respect the `sample_rate` corresponding to the `OutputFormat` passed in.
Adds
- `client.tts.sse` methods for generating audio using Server-Sent Events
- `client.tts.websocket` methods for managing a WebSocket connection and generating audio
- `client.tts.get_output_format()` to obtain `OutputFormat` object from output format name
- `client.tts.get_sample_rate()` to obtain `sample_rate` from output format name
- `client.voices.list()` to fetch a list of all available voices
- `client.voices.get()` to fetch a `VoiceMetadata` object from `voice_id`
- `client.voices.clone()` to clone a voice by specifying a filepath
- `client.voices.create()` to create a new voice
- Specifies `cartesia_version=2024-06-10` as default header for HTTP and WS requests
Removes
- `client.get_voices()`
- `client.get_voice_embedding()`
- `client.generate()`
- Ability to specify Numpy Array as a return type. We recommend using `np.frombuffer` with the appropriate `dtype`.
- Ability to clone voices using a link