#7272nd of 121 in VoiceFor voice, speech and music

audio8_tts

Compact zero-shot voice-cloning text-to-speech models, 0.6B and 0.1B

Releases and problems feed (RSS)

Stars
2.5k
License
Apache-2.0
Last commit
Nov 2026
Language
Python
Stars, last 16 weeks+2,336
30 Jul 2026: 173 stars6 Aug 2026: 397 stars13 Aug 2026: 517 stars20 Aug 2026: 673 stars27 Aug 2026: 1,007 stars3 Sept 2026: 1,282 stars10 Sept 2026: 1,404 stars17 Sept 2026: 1,505 stars24 Sept 2026: 1,545 stars1 Oct 2026: 1,572 stars8 Oct 2026: 1,597 stars15 Oct 2026: 1,778 stars22 Oct 2026: 1,894 stars29 Oct 2026: 1,991 stars5 Nov 2026: 2,366 stars12 Nov 2026: 2,448 stars19 Nov 2026: 2,509 stars

173 on 30 Jul 20262,509 on 19 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

Audio8 TTS provides two families of text-to-speech models with zero-shot voice cloning: 0.6B Realistic Air (23 languages) and 0.1B Preview (mainly Chinese and English). Inference runs through a Python script on a CUDA GPU, a llama.cpp/GGUF C++ runtime for Realistic Air, an ONNX INT8 runtime for CPU with the 0.1B model, or SGLang Omni with an OpenAI-compatible speech API.

Who it is for: Engineers needing self-hosted voice-cloning TTS on GPU, CPU or embedded targets

Strengths

  • Several runtimes: Python, llama.cpp/GGUF, ONNX INT8 on CPU, SGLang Omni
  • Realistic Air covers 23 languages and clones a voice from reference audio plus transcript
  • 0.6B models and repository code are Apache-2.0
  • Includes supervised fine-tuning guide and live listening demos

Weaknesses

  • 0.1B checkpoint uses a community license; companies at US$2M revenue or more need a paid license
  • No ONNX export for Realistic Air; ONNX INT4 is Preview 0.6B only
  • Per-language quality varies; benchmarks do not cover every supported language
  • macOS needs CPU (Metal experimental); Windows build steps untested

What it needs

  • GPU optional
  • Needs Python 3.10+, PyTorch (via requirements.txt), llama.cpp (patched, for GGUF), SGLang Omni (optional serving), ONNX Runtime (CPU path)
  • Models: Audio8-TTS-Realistic-Air 0.6B, Audio8-TTS-Preview 0.6B, Audio8-TTS-Preview 0.1B, GGUF F32, ONNX INT8

Known problems

Open now, reported in the last 90 days

Titles are the reporters' own words, taken from the project's GitHub issues and ranked by reactions. We have not tried to reproduce any of them.

Also in Voice

See all 121
Also in Voice
RankProjectScore
1voicestudioDesktop app for voice cloning, dubbing, dictation and audiobooks115.7k stars+35,283 (+44%) in 30 daysAGPL-3.098 out of 100
2handyOffline push-to-talk dictation app that pastes transcripts into any text field39.5k stars+5,163 (+15%) in 30 daysMIT93 out of 100
3whisper.cppC/C++ speech-to-text engine running OpenAI Whisper models locally55.4k stars+895 (+1.6%) in 30 daysMIT91 out of 100
#4meetilyDesktop app that records, transcribes and summarizes meetings locally Live demo ↗ (opens in a new tab)32.9k stars+999 (+3.1%) in 30 daysMIT88 out of 100
#5agentsPython framework for realtime voice and multimodal agents on LiveKit WebRTC Live demo ↗ (opens in a new tab)15.3k stars+421 (+2.8%) in 30 daysApache-2.086 out of 100