delayed-streams-modeling

Delayed Streams Modeling (DSM) is a flexible formulation for streaming, multimodal sequence-to-sequence learning.

Speech To Text

English only model

The main model handles english only, it has ~2.6b parameters.

PyTorch implementation

[Hugging Face]

# wget https://github.com/metavoiceio/metavoice-src/raw/main/assets/bria.mp3
python -m moshi.run_inference --hf-repo kyutai/stt-2.6b-en bria.mp3

MLX implementation

[Hugging Face]

# wget https://github.com/metavoiceio/metavoice-src/raw/main/assets/bria.mp3
python -m moshi_mlx.run_inference --hf-repo kyutai/stt-2.6b-en-mlx bria.mp3 --temp 0

Rust implementation

[Hugging Face]

The Rust implementation provides a server that can process multiple streaming queries in parallel. Dependening on the amount of memory on your GPU, you may have to adjust the batch size from the config file. For a L40S GPU, a batch size of 64 works well.

In order to run the server, install the moshi-server crate via the following command. The server code can be found in the kyutai-labs/moshi repository.

cargo install --features cuda moshi-server

Then the server can be started via the following command using the config file from this repository.

moshi-server worker --config configs/config-stt-hf.toml

Once the server has started you can run a streaming inference with the following script.

# wget https://github.com/metavoiceio/metavoice-src/raw/main/assets/bria.mp3
uv run scripts/asr-streaming-query.py bria.mp3

The script simulates some real-time processing of the audio. Faster processing can be triggered by setting the real-time factor, e.g. --rtf 500 will process the data as fast as possible.

English + French model

This model has ~1b parameters and supports both English and French.

PyTorch implementation

[Hugging Face]

# wget https://github.com/metavoiceio/metavoice-src/raw/main/assets/bria.mp3
python -m moshi.run_inference --hf-repo kyutai/stt-1b-en_fr bria.mp3

MLX implementation

[Hugging Face]

# wget https://github.com/metavoiceio/metavoice-src/raw/main/assets/bria.mp3
python -m moshi_mlx.run_inference --hf-repo kyutai/stt-1b-en_fr-mlx bria.mp3 --temp 0

Rust implementation

[Hugging Face]

The only difference with the en only model is the config file used when launching the server.

moshi-server worker --config configs/config-stt-enfr-hf.toml

Text To Speech

We're in the process of open-sourcing our TTS models. Check back for updates!

License

The present code is provided under the MIT license for the Python parts, and Apache license for the Rust backend. The web client code is provided under the MIT license. Note that parts of this code is based on AudioCraft, released under the MIT license.

The weights for the models are released under the CC-BY 4.0 license.