To speed up Long-context LLMs' inference, approximate and dynamic sparse calculate the attention, which reduces inference latency by up to 10x for pre-filling on an A100 while maintaining accuracy.

YAAPT Pitch Tracking function in PyTorch

Public code release associated with SceneScript.

Unofficial implementation

FlashSpeech: Efficient Zero-Shot Speech Synthesis

Diffusion-based singing voice pitch correction

Omni SenseVoice: High-Speed Speech Recognition with words timestamps 🗣️🎯

Computes the Energy Sliced Wasserstein Loss between two distributions. An optimal-transporty-energyish vibe distribution matching loss/regulariser.

PyTorch implementation of WaveFit [2022, Google] which is one of SOTA lightweight/fast speech vocoders.

phoneme tokenizer and grapheme-to-phoneme model for 8k languages

This is a text-processing frontend that converts graphemes to phonemes and then further converts those phonemes into articulatory features, for over 7000 languages.

16-fold memory access reduction with nearly no loss

The source code for the Interspeech 2024 paper "Lightweight Transducer Based on Frame Level Criterion".

Writing FLUX in Triton

A dialect of Lisp that's embedded in Python

[INTERSPEECH 2024] Official code for VoxSim: A perceptual voice similarity dataset

An all-purpose window upscaler for Windows 10/11.

openpilot is an operating system for robotics. Currently, it upgrades the driver assistance system in 275+ supported cars.

Ultra-low-bitrate Speech Codec for Speech Language Modeling Applications

an architecture for neural network inference in real-time audio applications

The reproduce training process for Moshi

EDM-HSE is an open audio dataset featuring 8000 house music drum loops.

Dataset and baseline code for the VocalSound dataset (ICASSP2022).

A very simple BERT implementation in PyTorch, which only depends on PyTorch itself.