GPA
Enrichment pending[AutoArk] GPA (General Purpose Audio) can do ASR, TTS and voice conversion with one tiny model!
GraphCanon updated today · GitHub synced today
Trust & integrity
Full report- Maintenance
- Steady (47d since push)
- As of today · Source: github_public_v1
- Provenance
- Not a fork · Organization account
- As of today · Source: github_public_v1
- Security (OSV)
- 34 low (34 low)
- As of today · Source: osv@v1
Public GitHub metadata and optional OSV dependency scans. Signals, not a guarantee. Trust methodology.
Overview
[AutoArk] GPA (General Purpose Audio) can do ASR, TTS and voice conversion with one tiny model!
Capability facts
- Languages
- python
Source: github.language+pyproject.toml · Jul 11, 2026
Categories
Tags
README
GPA: One Model for Speech Recognition, Text-to-Speech, and Voice Conversion
📢 Announcements
-
🚀 2026.04.29: GPA-v1.5 is here! GPA v1.5 Delivers near-SOTA TTS and ASR performance—in a single unified model. Start here →
-
🚀 2026.04.29: GPA-v1.5 ONNX Runtime is now available! Run ASR/TTS through ONNX CLI tools, a FastAPI service, or the browser UI with the new GPA-v1.5 ONNX runtime guide and runtime asset bundle.
-
🆕 2026.04.07: GPA-TTS FP16/FP32 Decoder — Higher-quality decoder options now available! For users with extra compute headroom, FP16 and FP32 SparkDetokenizer decoders are now available alongside INT8, delivering more stable and higher-quality speech synthesis. Selectable at runtime via CLI, API, or Web UI. Details →
-
📌 2026.03.31: GPA-TTS — Standalone lightweight TTS runtime released! Extracted from GPA with INT8/INT4 quantization for edge deployment. Among the smallest open-source TTS runtimes with voice cloning support! Details →
-
📚 GPA-v1.0 docs have moved. The original GPA-0.3B-preview quick start, deployment, benchmark, and evaluation pages now live in docs/GPA-v1.0.md.
All in one, built for all.
A single model delivering near-SOTA performance on TTS and ASR — fully unified, fully open!
📖 Abstract
GPA stands for General Purpose Audio.
A student’s GPA unifies performance across diverse subjects—from Calculus to Gym—into a single metric. Likewise, our GPA model integrates the three core audio tasks—TTS, ASR, and Voice Conversion—into one auto-regressive transformer.
GPA-v1.5 now delivers near-SOTA performance on ASR and TTS in a single unified model, with VC support on the roadmap.
Figure 1. GPA unifies speech understanding and generation in a single autoregressive audio-language model.
🗺️ Roadmap · 🚀 GPA-v1.5 Release · 🎙️ GPA-TTS · 🧭 GPA-v1.0 Archive · 📊 GPA-v1.5 Evaluation · 🔗 Citation
🗺️ Roadmap
| Category | Item | Status |
|---|---|---|
| Core Features | Unified LLM-based audio generation & understanding | ✅ |
| Native GPA-v1.5 Inference Pipeline | ✅ | |
| Native GPA-v1.5 Training Pipeline | ✅ | |
| GPA-v1.5 ONNX Runtime CLI/API/UI | ✅ | |
| GPA-v1.5 Interactive Demo | ⬜ | |
| GPA-v1.5 Basic Service Deployment (vLLM/FastAPI) | ⬜ | |
| Paper (ArXiv) | ✅ | |
| Model Releases | GPA-0.3B-preview | ✅ |
| GPA-v1.5 — major mainline release | ✅ | |
| GPA-TTS — Lightweight TTS runtime (INT8/FP16/FP32 + INT4 ONNX) | ✅ | |
| GPA-v1.5 Next Steps | Voice Conversion native path | ⬜ |
| Expanded deployment recipes | ⬜ | |
| Frameworks | torch | ✅ |
| vllm | ✅ | |
| llama-cpp | ✅ | |
| sglang | ✅ | |
| mlx-lm | ✅ | |
| rknn | ⬜ |