|
Download README.md from AXERA-TECH/LavaSR.AXERA: direct link, hf CLI and curl.
- Browser
- Download file 12.1 kB
-
https://huggingface.co/AXERA-TECH/LavaSR.AXERA/resolve/main/README.md
- Command line
-
hf download hf://AXERA-TECH/LavaSR.AXERA/README.md
-
curl -L -o README.md https://huggingface.co/AXERA-TECH/LavaSR.AXERA/resolve/main/README.md
12.1 kB
| library_name: axera | |
| pipeline_tag: audio-to-audio | |
| base_model: YatharthS/LavaSR | |
| language: | |
| - en | |
| - zh | |
| tags: | |
| - axera | |
| - ax650 | |
| - npu3 | |
| - lavasr | |
| - speech-super-resolution | |
| - audio-super-resolution | |
| - bandwidth-extension | |
| - speech-enhancement | |
| - speech-denoising | |
| - real-time | |
| - python | |
| - cpp | |
| # LavaSR v2 AX650 / AXERA | |
| **8–48 kHz 语音 → 48 kHz 超分 + 降噪 · AX650 NPU3 · C++ 全链路 RTF 0.061(约 16x realtime)· 3 个 axmodel 共 38 MB** | |
|   [](https://github.com/ZY-2012/LavaSR.AXERA) | |
| - GitHub: https://github.com/ZY-2012/LavaSR.AXERA | |
| - 上游: https://github.com/ysharma3501/LavaSR / https://huggingface.co/YatharthS/LavaSR | |
| > **AX650 板端实测说明**:AudioSR、LavaSR、FlowHigh 均已在 AX650N / NPU3 板端跑通。综合效果和速度最好的是 **LavaSR**:C++ 全链路 RTF ≈ 0.061(约 16× 实时),支持 8–48 kHz → 48 kHz 和降噪。**推荐优先使用 LavaSR。** | |
| > | |
| > **对比入口** | |
| > GitHub: [LavaSR](https://github.com/ZY-2012/LavaSR.AXERA) · [FlowHigh](https://github.com/ZY-2012/FlowHigh.AXERA) · [AudioSR](https://github.com/ZY-2012/AudioSR.AXERA) | |
| > HuggingFace: [LavaSR](https://huggingface.co/HY-2012/LavaSR.AXERA) · [FlowHigh](https://huggingface.co/HY-2012/FlowHigh.AXERA) · [AudioSR](https://huggingface.co/HY-2012/AudioSR.AXERA) | |
| ## 目录 | |
| - 快速开始 | |
| - 结果 | |
| - 推理速度 | |
| - 数据来源 | |
| - License | |
| ## 快速开始(AX650 板端) | |
| 从 HuggingFace 下载 `models/`、`bin/`、`python/`、`samples/`,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 --no-denoise | |
| # Python SDK | |
| bash run_ax650.sh samples/listen/8k/OSR_us_000_0010_input_8k.wav output.wav --python --no-denoise | |
| ``` | |
| 输入支持 8–48 kHz,输出固定 48 kHz。`--no-denoise` 仅做 BWE;去掉会同时跑 UL-UNAS 降噪。 | |
| ## 结果 | |
| ### 8 kHz 输入(Open Speech / Samplelib) | |
| #### English | |
| **OSR_us_000_0010** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_us_000_0010_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0010_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_us_000_0010_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0010_output_48k_spec.png" width="360"> | | |
| **OSR_us_000_0013** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_us_000_0013_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0013_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_us_000_0013_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0013_output_48k_spec.png" width="360"> | | |
| **OSR_us_000_0057** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_us_000_0057_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0057_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_us_000_0057_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0057_output_48k_spec.png" width="360"> | | |
| **OSR_us_000_0059** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_us_000_0059_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0059_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_us_000_0059_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0059_output_48k_spec.png" width="360"> | | |
| #### Chinese | |
| **OSR_cn_000_0072** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_cn_000_0072_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0072_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_cn_000_0072_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0072_output_48k_spec.png" width="360"> | | |
| **OSR_cn_000_0073** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_cn_000_0073_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0073_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_cn_000_0073_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0073_output_48k_spec.png" width="360"> | | |
| **OSR_cn_000_0074** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_cn_000_0074_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0074_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_cn_000_0074_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0074_output_48k_spec.png" width="360"> | | |
| **OSR_cn_000_0075** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_cn_000_0075_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0075_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/OSR_cn_000_0075_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0075_output_48k_spec.png" width="360"> | | |
| #### Samplelib(1 分钟,频谱图为前 10 s) | |
| **samplelib_speech_1m** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/samplelib_speech_1m_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/samplelib_speech_1m_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/samplelib_speech_1m_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/samplelib_speech_1m_output_48k_spec.png" width="360"> | | |
| ### 16 kHz 参照(人声) | |
| 同一批人声降采样到 16 kHz 后输入;8k 行为官方 8k 输入,16k 行为 16k 输入。 | |
| **speech_2** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/speech_2_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/speech_2_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/speech_2_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/speech_2_output_8k_48k_spec.png" width="360"> | | |
| | 16k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/16k/speech_2_input_16k.wav"></audio><br><img src="samples/listen/spectrograms/speech_2_input_16k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from16k/speech_2_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/speech_2_output_16k_48k_spec.png" width="360"> | | |
| **speech_3** | |
| | 版本 | 输入 | LavaSR 输出 | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/speech_3_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/speech_3_input_8k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from8k/speech_3_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/speech_3_output_8k_48k_spec.png" width="360"> | | |
| | 16k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/16k/speech_3_input_16k.wav"></audio><br><img src="samples/listen/spectrograms/speech_3_input_16k_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_from16k/speech_3_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/speech_3_output_16k_48k_spec.png" width="360"> | | |
| ### 带噪输入(降噪 + BWE) | |
| **OSR_us_000_0010 + 10 dB noise** | |
| | 版本 | 带噪输入 | LavaSR 输出(降噪 + BWE) | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_us_000_0010_noisy_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0010_noisy_input_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_noisy/OSR_us_000_0010_noisy_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_us_000_0010_noisy_output_spec.png" width="360"> | | |
| **OSR_cn_000_0072 + 10 dB noise** | |
| | 版本 | 带噪输入 | LavaSR 输出(降噪 + BWE) | | |
| |---|---|---| | |
| | 8k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/8k/OSR_cn_000_0072_noisy_input_8k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0072_noisy_input_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_noisy/OSR_cn_000_0072_noisy_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/OSR_cn_000_0072_noisy_output_spec.png" width="360"> | | |
| **speech_2 + 10 dB noise** | |
| | 版本 | 带噪输入 | LavaSR 输出(降噪 + BWE) | | |
| |---|---|---| | |
| | 16k | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/16k/speech_2_noisy_input_16k.wav"></audio><br><img src="samples/listen/spectrograms/speech_2_noisy_input_spec.png" width="360"> | <audio controls style="width:220px" src="https://huggingface.co/HY-2012/LavaSR.AXERA/resolve/main/samples/listen/48k_noisy/speech_2_noisy_output_48k.wav"></audio><br><img src="samples/listen/spectrograms/speech_2_noisy_output_spec.png" width="360"> | | |
| ## 推理速度:RTF | |
| | 链路 | Python RTF | C++ RTF | | |
| |---|---:|---:| | |
| | 降噪 + BWE | 0.278 | **0.061** | | |
| | 仅 BWE | 0.103 | **0.039** | | |
| ## 数据来源 | |
| - Open Speech Repository(English / Chinese 8k):http://voiptroubleshooter.com/open_speech/index.html | |
| - Samplelib speech 1 min:https://samplelib.com/mp3/sample-speech-1m.mp3 | |
| - 16 kHz 参照:公开人声样例经 48k->16k 降采样。 | |
| ## License | |
| Apache-2.0;KissFFT BSD-3-Clause。 | |