HN Showcase
Back to browse
GitHub Repository

An all-in-one, pure C++ inference engine for audio models, powered by ggml. Supports TTS, STT, VAD, voice conversion, music generation, and more, with highly optimized performance. No Python dependency.

483 starsC++

Run 25 Audio Models (TTS/STT/STS/Muisc) Locally in C++/GGML. No Python

by 0xshug0·Jul 10, 2026·2 points·1 comment
Visit ProjectView on HN

AI Analysis

●●●BangerWizardryZero to One

VibeVoice generates 94-minute podcasts 5x faster than real-time without Python.

Strengths
  • •CUDA paths run 1.8-5.0x faster than Python reference implementations
  • •Single ggml-based runtime replaces dozens of Conda environments
  • •25 model families including TTS, STT, voice conversion, music generation
Weaknesses
  • •Experimental JSON pipeline support not production-ready yet
  • •UI and API server flagged as needing community contributions
Category
AI/ML
Target Audience

ML engineers, audio developers, local AI enthusiasts

Similar To

llama.cpp · whisper.cpp · Coqui TTS

Similar Projects

Hardware●●●Banger

An Open-Source Yoto Toy with Qwen3-TTS

Voice cloning on ESP32 without cloud beats Yoto's subscription model completely.

WizardryZero to OneDark Horse
akadeb
313mo ago
AI/ML●●Solid

TTS.ai

Twenty-seven open-source TTS models in one UI with no signup required for the free tier.

SlickCrowd Pleaser
nadermx
302mo ago