Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Kyutai's Speech-To-Text and Text-To-Speech models based on the Delayed Streams Modeling framework.
| Date | Stars |
|---|---|
| 2026-07-31 | 2997 |
| 2026-08-02 | 3000 |
| 2026-08-03 | 3000 |
| 2026-08-06 | 3000 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Delayed Streams Modeling: Kyutai STT & TTS
This repo contains instructions and examples of how to run
[Kyutai Speech-To-Text](#kyutai-speech-to-text)
and [Kyutai Text-To-Speech](#kyutai-text-to-speech) models.
See also [Unmute](https://github.com/kyutai-labs/unmute), a voice AI system built using Kyutai STT and Kyutai TTS.
But wait, what is "Delayed Streams Modeling"? It is a technique for solving many streaming X-to-Y tasks (with X, Y in `{speech, text}`)
that formalize the approach we had with Moshi and Hibiki. See our [pre-print about DSM](https://arxiv.org/abs/2509.08753).
## Kyutai Speech-To-Text
<a href="https://huggingface.co/collections/kyutai/speech-to-text-685403682cf8a23ab9466886" target="_blank" style="margin: 2px;">
<img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-KyutaiSTT-blue" style="display: inline-block; vertical-align: middle;"/>
</a>
<a target="_blank" href="https://colab.research.google.com/github/kyutai-labs/delayed-streams-modeling/blob/main/stt_pytorch.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
**More details can be found on the [project page](https://kyutai.org/next/stt).**
Kyutai STT models are optimized for real-time usage, can be batched for efficiency, and return word level timestamps.
We provide two models:
- `kyutai/stt-1b-en_fr`, an English and French model with ~1B parameters, a 0.5 second delay, and a [semantic VAD](https://kyutai.org/next/stt#semantic-vad).
- `kyutai/stt-2.6b-en`, an English-only model with ~2.6B parameters and a 2.5 second delay.
These speech-to-text models have several advantages:
- Streaming inference: the models can process audio in chunks, which allows
for real-time transcription, and is great for interactive applications.
- Easy batching for maximum efficiency: a H100 can process 400 streams in
real-time.
- They return word-level timestamps.
- The 1B model has a semantic Voice Activity Detection (VAD) component that
can be used to detect when the user is speaking. This is especially useful
for building voice agents.
### Implementations overview
We provide different implementations of Kyutai STT for different use cases.
Here is how to choose which one to use:
- **PyTorch: for research and tinkering.**
If you want to call the model from Python for research or experimentation, use our PyTorch implementation.
- **Rust: for production.**
If you want to serve Kyutai STT in a production setting, use our Rust server.
Our robust Rust server provides streaming access to the model over websockets.
We use this server to run [Unmute](https://unmute.sh/); on a L40S GPU, we can serve 64 simultaneous connections at a real-time factor of 3x.
- **MLX: for on-device inference on iPhone and Mac.**
MLX is Apple's ML framework that allows you to use hardware acceleration on Apple silicon.
If you want to run the model on a Mac or an iPhone, choose the MLX implementation.
<details>
<summary>PyTorch implementation</summary>
<a href="https://huggingface.co/kyutai/stt-2.6b-en" target="_blank" style="margin: 2px;">
<img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue" style="display: inline-block; vertical-align: middle;"/>
</a>
<a target="_blank" href="https://colab.research.google.com/github/kyutai-labs/delayed-streams-modeling/blob/main/stt_pytorch.ipynb">
<img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>
For an example of how to use the model in a way where you can directly stream in PyTorch tensors,
[see our Colab notebook](https://colab.research.google.com/github/kyutai-labs/delayed-streams-modeling/blob/main/stt_pytorch.ipynb).
This requires the [moshi package](https://pypi.org/project/moshi/)
with version 0.2.6 or later, which can be installed via pip.
If you just want to run the model on a file, you can use `moshi.run_inference`.
```bash
python -m moshi.Excerpt of 15,385 characters
Read on GitHub
Laurent Mazare · France
74
Václav Volhejn · @kyutai-labs · France
15
Alexandre Défossez · France
5
Eugene Kharitonov · France
2
1
1
1
1
Gabriel de Marmiesse · Kyutai · France
1
1
Chenghao Mou · Ireland
1
Bai Li
1
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:edaed81da1ea10e2, desc:speech-to-text, desc:text-to-speech