Hibiki is a model for streaming speech translation (also known as simultaneous translation). Unlike offline translation—where one waits for the end of the source utterance to start translating--- H…

Make text LLMs listen and speak

Kyutai's Speech-To-Text and Text-To-Speech models based on the Delayed Streams Modeling framework.

Simple high-throughput inference library

Experimental compiler for deep learning models

Fine-tuning Moshi/J-Moshi on your own spoken dialogue data

Moshi is a speech-text foundation model and full-duplex spoken dialogue framework. It uses Mimi, a state-of-the-art streaming neural audio codec.

Googles NotebookLM but local

python bindings for symphonia/opus - read various audio formats from python and write opus files

Kyutai with an "eye"

Proof of concept for running moshi/hibiki using webrtc

J-Moshi: A Japanese Full-duplex Spoken Dialogue System

Implementation of the paper "Lossless Compression of Vector IDs for Approximate Nearest Neighbor Search" by Severo et al.

A small rust-based data loader

A tool to reconstruct Capcom CP-1 paper sheets

protein utilities in Rust

A Fish Speech implementation in Rust, with Candle.rs

A voice chat app

implement llava using candle

A text embedding extension for the Polars Dataframe library.

Implementing the BitNet model in Rust

An unofficial implementation of BitNet

A Keras like abstraction layer on top of the Rust ML framework candle