#3131st of 38 in VoiceFor voice, speech and music

cosyvoice

Zero-shot multilingual text-to-speech with voice cloning and streaming output

Stars
24.1k
License
Apache-2.0
Last commit
May 2026
Language
Python
Stars, last 29 weeks+4,023
31 Mar 2026: 20,060 stars7 Apr 2026: 20,185 stars14 Apr 2026: 20,336 stars21 Apr 2026: 20,464 stars28 Apr 2026: 20,569 stars5 May 2026: 20,671 stars12 May 2026: 20,808 stars19 May 2026: 20,938 stars26 May 2026: 21,085 stars2 Jun 2026: 21,238 stars9 Jun 2026: 21,395 stars16 Jun 2026: 21,530 stars23 Jun 2026: 21,661 stars30 Jun 2026: 21,768 stars7 Jul 2026: 21,897 stars14 Jul 2026: 22,049 stars21 Jul 2026: 22,219 stars28 Jul 2026: 22,363 stars4 Aug 2026: 22,477 stars11 Aug 2026: 22,612 stars18 Aug 2026: 22,733 stars25 Aug 2026: 22,843 stars1 Sept 2026: 23,341 stars8 Sept 2026: 23,464 stars15 Sept 2026: 23,584 stars22 Sept 2026: 23,685 stars29 Sept 2026: 23,774 stars6 Oct 2026: 23,836 stars13 Oct 2026: 23,938 stars20 Oct 2026: 24,083 stars

20,060 on 31 Mar 202624,083 on 20 Oct 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

CosyVoice is an LLM-based text-to-speech system from the FunAudioLLM team. Fun-CosyVoice 3.0 (0.5B) clones voices zero-shot across 9 languages and 18+ Chinese dialects, and supports text-in and audio-out streaming with latency as low as 150ms. The repo includes inference and training scripts, a web demo, gRPC and FastAPI servers, and vLLM and TensorRT-LLM runtimes.

Who it is for: Engineers self-hosting multilingual TTS and voice cloning

Strengths

  • Covers 9 languages and 18+ Chinese dialects, with cross-lingual zero-shot voice cloning
  • Bi-streaming (text in, audio out) with latency as low as 150ms
  • Pinyin and CMU phoneme inpainting; instructions for emotion, speed and volume
  • Ships training scripts, FastAPI and gRPC servers, plus vLLM and TensorRT-LLM runtimes

Weaknesses

  • Setup is manual: conda env, git submodules, separate model downloads, sox install
  • README gives no RAM or VRAM requirements
  • vLLM supported only on 0.9.0 and 0.11.x+; versions in between are untested
  • No tagged releases; install commands default to Chinese PyPI mirrors

What it needs

  • Docker
  • Needs Python 3.10, conda, sox, vLLM (optional), TensorRT-LLM (optional), ttsfrd (optional)
  • Models: Fun-CosyVoice3-0.5B-2512, CosyVoice2-0.5B, CosyVoice-300M, CosyVoice-300M-SFT, CosyVoice-300M-Instruct
  • port 50000

Also in Voice

See all 38
Also in Voice
RankProjectScore
1voicestudioDesktop app for voice cloning, dubbing and dictation with local speech models79.5k starsAGPL-3.083 out of 100
2whisper.cppC/C++ port of OpenAI Whisper for local speech-to-text transcription54.5k starsMIT77 out of 100
3VoiceboxLocal voice studio for cloning, TTS, dictation and agent speech58.1k starsMIT71 out of 100
#4buzzOffline desktop app for transcribing and translating audio with Whisper22.1k starsMIT69 out of 100
#5meetilyDesktop app that records, transcribes and summarizes meetings locally Live demo ↗ (opens in a new tab)32k starsMIT68 out of 100