Downloads · 30 days
17
2% of all-time downloads
kokoro.best
inoryQwQ/kokoro.best
kokoro.best is a text-to-speech model from inoryQwQ. Use it when you need text read aloud. It is set up for kokoro-axera. The card lists the license as apache-2.0.
Kokoro-82M 文本转语音模型在 AX650 NPU 上的部署方案。3 模型全 NPU 推理，支持 Python/C++ SDK，中英文前端。
All-time downloads
733
Public
Repo size
890 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.axmodel187 MB · 42%
At a glance
- Task
- Text-to-Speech
- Library
- kokoro-axera
- License
- apache-2.0
- Access
- Public
- Created
- Jul 22, 2026
- Updated
- Jul 24, 2026
- SHA
- 2bc58f96
Datasets
Model card
From the Hugging Face model README
Kokoro TTS for AX650 NPU
Kokoro-82M 文本转语音模型在 AX650 NPU 上的部署方案。3 模型全 NPU 推理，支持 Python/C++ SDK，中英文前端。
管线架构
文本 → espeak-ng/Jieba前端 → Encoder(U16 NPU) → F0N(Mixed FP32 NPU) → Decoder(S16 NPU) → HAR+ISTFT(CPU) → 24kHz音频
| 模型 | 量化 | 大小 | 耗时 |
|---|---|---|---|
| Encoder | AXERA U16 | 96MB | 31ms |
| F0N | Mixed FP32 | 14MB | 8ms |
| Decoder | S16 | 70MB | 102ms |
性能
| 指标 | Python SDK | C++ SDK |
|---|---|---|
| RTF | 0.10 | — |
| EN ASR (10短语) | 9/10 | 7/10 |
| CN 长句 ASR | — | ✅ |
快速开始
Python SDK
cd python && pip install -r requirements.txt
# 安装 pyaxengine
git clone https://github.com/AXERA-TECH/pyaxengine.git && pip install -e pyaxengine/
# CLI
python -m kokoro_axera.cli -t "həlˈoʊ wˈɜːld" -o hello.wav
# OpenAI server
python -m kokoro_axera.server --port 8880
C++ SDK
export LD_PRELOAD='/soc/lib/libax_sys.so /soc/lib/libax_engine.so'
export LD_LIBRARY_PATH=/soc/lib:./cpp/lib
# English
./cpp/bin/kokoro_cli ./models ./models/espeak-ng-data "Hello world" output.wav en
# Chinese
./cpp/bin/kokoro_cli ./models ./models/espeak-ng-data "你好世界" output.wav zh
已知限制
- 仅支持 af_heart 单说话人
- Decoder S16 极短句有残余失真
从零复现
- Task
- Text-to-Speech
- Library
- kokoro-axera
- License
- apache-2.0
- Languages
- en
- Created
- Jul 22, 2026
- Updated
- Jul 24, 2026