Higgs Audio Studio — Pinokio launcher

One-click cross-platform installer for Higgs Audio Studio — Higgs Audio v3 TTS: expressive speech in 100+ languages, voice cloning, AI text director, podcast & audiobook. 100% offline.

This repository is the Pinokio launcher for Higgs Audio Studio — the actual app lives there. This repo only contains the scripts Pinokio runs to install, start, update and reset the app in an isolated, cross-platform environment.

Install

1. Download and install Pinokio.
2. Open in Pinokio:
   • 🚀 Install (1-click) — direct install URL
   • 📂 Browse app page — catalog page on beta.pinokio.co
3. Click Install inside Pinokio, then Start. The first start is longer (one-time torch.compile warm-up); later runs are ~2× faster.

What this launcher does

• Isolated Python 3.12 venv via uv — no system-wide installs
• PyTorch auto-selected by GPU/OS — CUDA 12.8 (NVIDIA x64), CUDA 13.0 (NVIDIA aarch64), DirectML (AMD Win), ROCm 6.3 (AMD Linux), MPS/CPU (macOS), CPU fallback
• Triton for torch.compile (~2× on NVIDIA bf16). Higgs uses PyTorch SDPA flash kernels — external Flash-Attention 2 is not required
• llama-cpp-python for the GGUF AI text-director on GPU (cu124) or CPU
• CUDA runtime DLLs copied next to llama.dll on Windows NVIDIA (copy_cuda_dlls.py)
• Starter voice pack (~51 RU/EN presets) downloaded automatically; 743 extra Russian voices on-demand in-app
• Bundled Node.js + CUDA from Pinokio's ai bundle — no separate downloads
• Gradio auto-picks a free port via kernel.port(); NO_AUTO_BROWSER=true prevents a duplicate browser tab
• Env isolation: HF_HOME, TRANSFORMERS_CACHE, TORCH_HOME all point inside the launcher folder

Launch modes

Menu item	What it runs
Start	python app.py — Gradio on an auto-assigned port, full 6-tab UI (TTS / Expression + Director / Cloning / Podcast / Audiobook / Batch)
Update	git pull the app, then re-run install (picks up new dependencies)
Save Disk Space	Dedup venv libraries via fs.link
Reset	Wipe app/env to reinstall from scratch

Platform support matrix

OS	GPU	Status	Acceleration
Windows 10/11	NVIDIA RTX 30xx–50xx	✅ tested	CUDA 12.8 + Triton + GGUF on GPU
Windows 10/11	NVIDIA RTX 20xx	✅ expected	CUDA 12.8 + Triton
Linux x64	NVIDIA RTX 20xx–50xx	✅ expected	CUDA 12.8 + Triton
Linux aarch64	NVIDIA DGX Spark / Jetson	✅ expected	CUDA 13.0
Windows	AMD RDNA3+	✅ expected	DirectML (director on CPU)
Linux	AMD RDNA3+	✅ expected	ROCm 6.3 (director on CPU)
macOS	Apple Silicon M1–M4	✅ expected	MPS (director on CPU)
macOS	Intel	⚠️ CPU only	torch CPU
Any	CPU only	⚠️ very slow	CPU

Higgs v3 uses PyTorch SDPA (flash kernels built in); it does not support external Flash-Attention 2. torch.compile adds ~2× on NVIDIA.

Features (inside the app)

• 🎙️ TTS — 100+ languages, expressive control via inline tags, ⏹ Stop on the fly, output WAV / MP3 / FLAC / OGG
• 🎭 Expression + 🤖 AI director — a local GGUF LLM (Qwen3.5-9B) normalizes text and inserts emotion/prosody/sound tags by meaning
• 🧬 Voice cloning — zero-shot from a reference clip with auto-transcription
• 🎬 Podcast — multi-speaker dialogue with loudness leveling across speakers (LUFS −16)
• 📚 Audiobook — narrator + characters, consistent voices, long-form
• 📦 Batch + auto-enrich everywhere · RU / EN UI

Links

• App: https://github.com/timoncool/HiggsAudio-Studio
• Base model: bosonai/higgs-audio-v3-tts-4b
• Pinokio: https://pinokio.co

Authors

• Nerual Dreming — Telegram | neuro-cartel.com | ArtGeneration.me
• Нейро-Софт — Telegram | portable AI builds

Support the Author

I build open-source software and do AI research. Most of what I create is free and available to everyone. Your donations help me keep creating without worrying about where the next meal comes from =)

All donation methods | dalink.to/nerual_dreming | boosty.to/neuro_art

• BTC: 1E7dHL22RpyhJGVpcvKdbyZgksSYkYeEBC
• ETH (ERC20): 0xb5db65adf478983186d4897ba92fe2c25c594a0c
• USDT (TRC20): TQST9Lp2TjK6FiVkn4fwfGUee7NmkxEE7C

Star History