0xShugo

Private, secure, edge-native AI. Building real-time inference, speech systems, and AI that runs on real devices.

[0xShugo @0xShug0]

One month ago, audio.cpp was open-sourced. Yesterday, audio.cpp reached 1,000 GitHub stars and Alibaba Tongyi Lab integrated audio.cpp into the FunASR deployment platform🎉.

In one month:
📷 The project grew from 12 to 44 model families, with 31 contributors.
📷 Engineers from Alibaba and Mistral AI contributed models and improvements to make them more production-ready.
📷 Shared model specifications, reusable components, CLI and server interfaces, streaming support, and experimental pipelines moved the project closer to the original vision of a unified AI runtime.

Thank you to everyone who has contributed, tested, reported issues, or simply given the project a star. Here's to the next milestone. 📷🎉🚀

[0xShugo @0xShug0]

@shinshin86 Check our C++ version of Omnivoice (along with 40+ other audio models in one runtime). No Python and 1.8x to 8x faster than Python on CUDA! https://t.co/RpQznHV5Cw

[0xShugo @0xShug0]

https://t.co/rMGYINytcn

[0xShugo @0xShug0]

@C_of_Creativity Thanks for including the benchmark in the repo! I’m working on porting text-to-audio to audio.cpp, with text-to-video as a free bonus 😄. This is exactly the kind of reproducible performance baseline I’ve been looking for.

[0xShugo @0xShug0]

@simplifyinAI Check our c++/ggml version of pocket-tts (and 40+ other audio models in one runtime)! https://t.co/RpQznHV5Cw

[0xShugo @0xShug0]

Nice work! In case anyone is interested in a local version, audio.cpp is an all-in-one AI audio runtime that supports 40+ state-of-the-art audio models across TTS, ASR, music generation, and more, with highly optimized performance. A new UI is also open for testing now! https://t.co/K9H1vbstsx

[0xShugo @0xShug0]

One month ago, audio.cpp was open-sourced. Yesterday, audio.cpp reached 1,000 GitHub stars and Alibaba Tongyi Lab integrated audio.cpp into the FunASR deployment platform🎉.

In one month:
📷 The project grew from 12 to 44 model families, with 31 contributors.
📷 Engineers from Alibaba and Mistral AI contributed models and improvements to make them more production-ready.
📷 Shared model specifications, reusable components, CLI and server interfaces, streaming support, and experimental pipelines moved the project closer to the original vision of a unified AI runtime.

Thank you to everyone who has contributed, tested, reported issues, or simply given the project a star. Here's to the next milestone. 📷🎉🚀

[0xShugo @0xShug0]

VibeVoice 1.5B running locally...On an iPhone! Only ~2.2 GB of memory and up to 1.28× faster than realtime.

[0xShugo @0xShug0]

@OpenAI Vibevoice 1.5B running locally on an iPhone! 2.2GB VRAM and 1.2x faster than realtime.

[0xShugo @0xShug0]

Star our repo so you don’t miss future updates! The model was uploaded to the audio.cpp HF repo. I will upload the xcframework later, and then push the code to a branch after release 0.6.

[0xShugo @0xShug0]

VibeVoice 1.5B running locally...On an iPhone! Only ~2.2 GB of memory and up to 1.28× faster than realtime.

[0xShugo @0xShug0]

Another longform demo

[0xShugo @0xShug0]

We’re working on a C++/GGML implementation of Muscriptor in audio.cpp! It should be easy to port to other platforms, including iOS. If you’re a fan of music models, audio.cpp already supports ACE-Step, HeartMuLa, Stable Audio, and more. Follow our repo or X account so you don’t miss the update!

[0xShugo @0xShug0]

Very cool work. This is somewhat aligned with audio.cpp’s goal: running SoTA audio models locally at the edge to keep audio workflows responsive and reduce reliance on cloud compute. With deep optimization, complex audio workflows can run efficiently at the edge. I’m going to share a demo of VibeVoice 1.5B running directly on iOS at faster-than-real-time speed.

[0xShugo @0xShug0]

@AILogDev audio.cpp released a C++/GGML version of irodoriTTS v4. https://t.co/RpQznHV5Cw. 28x faster than realtime on RTX 5090.

[0xShugo @0xShug0]

@umiyuki_ai audio.cpp now supports Irodori-TTS-v4-Small (TTS, voice clone and voice design)! 28x faster than realtime with Q8 GGUF on RTX 5090.

Try at https://t.co/RpQznHVDs4

[0xShugo @0xShug0]

@rotejin audio.cpp now supports Irodori-TTS-v4-Small (TTS, voice clone and voice design)! 28x faster than realtime with Q8 GGUF on RTX 5090.

Try at https://t.co/RpQznHV5Cw

[0xShugo @0xShug0]

@rootport audio.cpp now supports Irodori-TTS-v4-Small (TTS, voice clone and voice design)! 28x faster than realtime with Q8 GGUF on RTX 5090.

Try at https://t.co/RpQznHV5Cw

[0xShugo @0xShug0]

@rotejin audio.cpp now supports Irodori-TTS-v4-Small! 28x faster than realtime with Q8 GGUF on RTX 5090.

Try at https://t.co/RpQznHV5Cw

[0xShugo @0xShug0]

@JPAI_HEAVEN audio.cpp now supports Irodori-TTS-v4-Small! 28x faster than realtime with Q8 GGUF on RTX 5090.

Try at https://t.co/RpQznHV5Cw

[0xShugo @0xShug0]

@ivy432hz audio.cpp now supports Irodori-TTS-v4-Small! 28x faster than realtime with Q8 GGUF on RTX 5090.

Try at https://t.co/RpQznHV5Cw