FastWave.AXERA / README.md
HY-2012's picture
docs: build-from-source section
c7e7722 verified
|
Raw History Blame Contribute Delete
4.98 kB
---
license: apache-2.0
language:
- en
- zh
tags:
- axera
- ax650
- npu
- audio-super-resolution
- speech-super-resolution
- bandwidth-extension
- audio-to-audio
library_name: axera
pipeline_tag: audio-to-audio
base_model: Nikait/FastWave
---
# FastWave.AXERA
**AX650 上的通用音频超分:6–48 kHz → 48 kHz。C++ 全链路 RTF 10.97(8 步)/ 5.51(4 步),Python SDK RTF 10.96 / 5.49;单 axmodel 11.9 MB,Python/C++ SDK 即取即用。**
![RTF](https://img.shields.io/badge/RTF%20%28C%2B%2B%29-10.97-brightgreen) ![Size](https://img.shields.io/badge/axmodel-11.9%20MB-green) ![License](https://img.shields.io/badge/license-Apache--2.0-blue)
- HuggingFace: https://huggingface.co/HY-2012/FastWave.AXERA
- GitHub: https://github.com/ZY-2012/FastWave.AXERA
- 上游: https://github.com/Nikait/FastWave
## 目录
- 快速开始
- 结果
- 推理速度
- 文件说明
- 数据来源
- License
## 快速开始(AX650 板端)
本仓库即完整包(`models/` + `bin/` + `python/` + `samples/`),下载后在包根目录运行:
```bash
# C++(默认)
bash run_ax650.sh samples/listen/8k/OSR_us_000_0010_input_8k.wav output.wav
# Python SDK
bash run_ax650.sh samples/listen/8k/OSR_us_000_0010_input_8k.wav output.wav --python
# 4 步采样(更快)/ 打开后置降噪
bash run_ax650.sh samples/listen/16k/speech_2_input_16k.wav output.wav --steps 4
bash run_ax650.sh samples/listen/8k/OSR_cn_000_0072_input_8k.wav output.wav --denoise
```
输入支持 6–48 kHz,输出固定 48 kHz。
## 结果
### 8 kHz 输入(Open Speech Repository)
| 版本 | 输入 | FastWave.AXERA 输出 |
|---|---|---|
| **OSR_us_000_0010** | [▶](samples/listen/8k/OSR_us_000_0010_input_8k.wav)<br><img src="samples/listen/spectrograms/OSR_us_000_0010_input_spec.png" width="360"> | [▶](samples/listen/48k_from8k/OSR_us_000_0010_output_48k.wav)<br><img src="samples/listen/spectrograms/OSR_us_000_0010_output_spec.png" width="360"> |
| **OSR_us_000_0013** | [▶](samples/listen/8k/OSR_us_000_0013_input_8k.wav)<br><img src="samples/listen/spectrograms/OSR_us_000_0013_input_spec.png" width="360"> | [▶](samples/listen/48k_from8k/OSR_us_000_0013_output_48k.wav)<br><img src="samples/listen/spectrograms/OSR_us_000_0013_output_spec.png" width="360"> |
| **OSR_cn_000_0072** | [▶](samples/listen/8k/OSR_cn_000_0072_input_8k.wav)<br><img src="samples/listen/spectrograms/OSR_cn_000_0072_input_spec.png" width="360"> | [▶](samples/listen/48k_from8k/OSR_cn_000_0072_output_48k.wav)<br><img src="samples/listen/spectrograms/OSR_cn_000_0072_output_spec.png" width="360"> |
| **OSR_cn_000_0074** | [▶](samples/listen/8k/OSR_cn_000_0074_input_8k.wav)<br><img src="samples/listen/spectrograms/OSR_cn_000_0074_input_spec.png" width="360"> | [▶](samples/listen/48k_from8k/OSR_cn_000_0074_output_48k.wav)<br><img src="samples/listen/spectrograms/OSR_cn_000_0074_output_spec.png" width="360"> |
### 16 kHz 输入(人声)
| 版本 | 输入 | FastWave.AXERA 输出 |
|---|---|---|
| **speech_2** | [▶](samples/listen/16k/speech_2_input_16k.wav)<br><img src="samples/listen/spectrograms/speech_2_input_spec.png" width="360"> | [▶](samples/listen/48k_from16k/speech_2_output_48k.wav)<br><img src="samples/listen/spectrograms/speech_2_output_spec.png" width="360"> |
| **speech_3** | [▶](samples/listen/16k/speech_3_input_16k.wav)<br><img src="samples/listen/spectrograms/speech_3_input_spec.png" width="360"> | [▶](samples/listen/48k_from16k/speech_3_output_48k.wav)<br><img src="samples/listen/spectrograms/speech_3_output_spec.png" width="360"> |
### 8k → 48k 频谱对比(上:输入,下:输出)
<img src="samples/listen/spectrograms/OSR_us_000_0010.png" width="720">
## 推理速度:RTF
AX650(NPU3,BSP V3.10.2),chunk 32768 点(0.68 s):
| 链路 | 8 步 | 4 步 | 无重叠(8 步) |
|---|---:|---:|---:|
| C++ | 10.97 | 5.51 | 6.29 |
| Python SDK | 10.96 | 5.49 | — |
## 文件说明
| 路径 | 内容 |
|---|---|
| `models/fastwave_nfe.axmodel` | 单次 NFE 图(11.9 MB,U16 激活 / S8 权重 / U16 输出) |
| `models/assets/*.npy` | 扩散 embedding 权重(host 端计算) |
| `bin/fastwave_ax650` | C++ 推理程序(aarch64,AX650) |
| `python/` | Python SDK(numpy + axengine) |
| `samples/` | 8k / 16k 输入与 48k 输出、频谱图 |
| `reports/` | 精度与性能报告 |
## 从源码编译(C++)
```bash
bash cpp/download_toolchains.sh # Arm GNU 9.2 + AX650 BSP SDK(约 1.1 GB,仅一次)
bash cpp/build_ax650.sh # -> bin/fastwave_ax650
```
`build_ax650.sh` 在缺少工具链/BSP 时会自动调用 `download_toolchains.sh`;已有工具链可用
`TOOLCHAIN_ROOT=` / `BSP_MSP_DIR=` 覆盖。
## 数据来源
- Open Speech Repository(English / Chinese 8k):http://voiptroubleshooter.com/open_speech/index.html
- 16 kHz 参照:公开人声样例经 48k → 16k 降采样
- 上游模型权重:Nikait/FastWave(MIT)
## License
Apache-2.0。