audio

huggingface/transformers

A unified model-definition framework spanning text, vision and audio — Transformers acts as the shared definition layer for models like DeepSeek, Gemma, GLM and Qwen, supporting both inference and training across text, vision, audio and multimodal tasks. It suits researchers and engineers who want one consistent API and access to pretrained checkpoints from the Hub instead of per-architecture code.

🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

OpenBMB/VoxCPM

VoxCPM2 brings tokenizer-free TTS with voice design and cloning — VoxCPM2 drops the usual discrete tokenizer step for speech synthesis, aiming at more natural multilingual TTS. It supports creative voice design and realistic voice cloning, with live demo, docs, and model weights on Hugging Face/ModelScope. Useful for developers and researchers building speech apps who want flexible, high-fidelity voice generation.

VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning

alyssaxuu/screenity

A Free Chrome Screen Recorder Built Around Privacy — Screenity is a Chrome extension for screen recording and annotation with no usage caps, aimed at anyone making tutorials, demos, or feedback videos without paying or sending footage to a third party. It uses local tools like ffmpeg and TensorFlow, and an optional EU-hosted Pro tier adds link sharing, multi-scene editing, and captions.

The free and privacy-friendly screen recorder with no limits 🎥

pion/webrtc

A native Go toolkit for building WebRTC apps without C bindings — Pion WebRTC lets Go developers build real-time audio, video and data applications without wrapping C/C++ libraries, and it compiles to WebAssembly for browser use too. Useful for anyone building P2P streaming, VoIP or live-video tools who wants a memory-safe, Go-native stack instead of libwebrtc bindings.

Pure Go implementation of the WebRTC API

crmne/spotifast

A Rust-native Spotify client built for speed, not a browser shell — Spotifast is a cross-platform Spotify client written in Rust with egui, using librespot for playback instead of a bundled browser engine. It typically uses 100-250MB RAM, starts in under a second, and ships with full translations for all supported languages. Playback requires Spotify Premium; free accounts can only browse and search.

Spotify, native and fast. One lightweight Rust app for your whole library, local playback, and Spotify Connect on Linux, macOS, and Windows.

Leonxlnx/earshot

Giving coding agents the ability to actually hear sound — earshot lets AI coding agents analyze, describe and judge audio—matching sounds to descriptions, detecting sync issues, telling real recordings from synthesized ones. v0.3 adds a 201-effect code-rendered sound bank for scoring and syncing film audio, all CPU-only via CLI, Python API or MCP server.

Ears for coding agents: let AI hear, review and make sound. CPU-only, open source.

tphakala/birdnet-go

24/7 local AI bird, bat, and wildlife sound ID, no cloud needed — A self-hosted, written-in-Go rewrite-style alternative to BirdNET-Pi, running 24/7 on devices as small as a Raspberry Pi. It supports multiple local AI models, including bat detection, for continuous soundscape monitoring without cloud dependency, aimed at hobbyist naturalists and citizen scientists.

Self-hosted realtime soundscape analyser for birds, bats and other wildlife. Multi-model local AI inference, runs 24/7 on a Raspberry Pi.

mardous/BoomingMusic

Clean, fast, and Material-driven Android music player with powerful features.

dsmurrell/onkey

A Mac singing trainer that learns melodies by just listening — OnKey listens to whatever plays in Spotify, Apple Music, or YouTube, separates the vocal, and teaches itself the melody after one play through. Later plays sync live from the audio alone, scrolling notes show your pitch against the original, and you get a per-line score. It's a free, open-source, on-device alternative to subscription karaoke apps, built for Apple Silicon Macs.

Learn to sing the songs you love. A free, open-source singing trainer for Mac that listens to the music you play.

kvoltmer/Audionaut

A lightweight multitrack editor that AI agents can drive directly — Audionaut is a free, open-source C++/JUCE multitrack audio editor for music, podcasts and multitrack recordings, lighter than a full DAW. It exposes MCP, letting Claude or other AI agents perform cuts, crossfades and track edits as single undo steps while you work alongside it in the app.

Audionaut professional audio editing and audio recording

andrewfader/blokkily

A pattern-first DAW that can be driven by an AI agent over MCP — Blokkily lets you edit one pattern across tracker, piano roll or step grid views, then arrange it across tracks with CLAP/VST3 plugins, SoundFonts and built-in samplers. Notably, it supports composing via an MCP-driven agent, as shown building a chiptune track with CLAP synths YANES and FUZZBOI.

An experimental pattern-first DAW