Fix HF README: add YAML frontmatter, update tree/C++ section to prebuilt-only layout
Browse files
README.md
CHANGED
|
@@ -1,3 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# openWakeWord.AXERA
|
| 2 |
|
| 3 |
[openWakeWord](https://github.com/dscripka/openWakeWord) 语音唤醒 AXERA 推理 demo。
|
|
@@ -5,7 +18,7 @@
|
|
| 5 |
## 功能
|
| 6 |
|
| 7 |
- 支持 AX650 和 AX630C;
|
| 8 |
-
- 支持 Python 和 C++
|
| 9 |
|
| 10 |
## 目录结构
|
| 11 |
|
|
@@ -16,20 +29,25 @@ OpenWakeWord.AXERA/
|
|
| 16 |
│ └── 630C/ # AX630C axmodel
|
| 17 |
├── config/ # STFT/mel 权重
|
| 18 |
├── audio/openwakeword/
|
| 19 |
-
├── scripts/ # Python
|
| 20 |
-
├── cpp/
|
|
|
|
|
|
|
| 21 |
└── README.md
|
| 22 |
```
|
| 23 |
|
|
|
|
|
|
|
|
|
|
| 24 |
|
| 25 |
## Python
|
| 26 |
|
| 27 |
```bash
|
| 28 |
-
创建虚拟环境
|
| 29 |
conda create -n OpenWakeWord python=3.10
|
| 30 |
conda activate OpenWakeWord
|
| 31 |
|
| 32 |
-
安装
|
| 33 |
pip install axengine-x.x.x-py3-none-any.whl
|
| 34 |
|
| 35 |
pip install -r requirements.txt
|
|
@@ -56,67 +74,38 @@ python scripts/openwakeword_ax630c.py \
|
|
| 56 |
--output outputs/openwakeword_ax630c.json
|
| 57 |
```
|
| 58 |
|
| 59 |
-
## C++
|
| 60 |
-
|
| 61 |
-
### 编译环境
|
| 62 |
-
|
| 63 |
-
AX650 和 AX630C 共用 Arm GNU 9.2 AArch64 交叉编译器:
|
| 64 |
-
|
| 65 |
-
<https://developer.arm.com/-/media/Files/downloads/gnu-a/9.2-2019.12/binrel/gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu.tar.xz>
|
| 66 |
-
|
| 67 |
-
```bash
|
| 68 |
-
mkdir -p cpp/toolchains
|
| 69 |
-
wget -P cpp/toolchains \
|
| 70 |
-
https://developer.arm.com/-/media/Files/downloads/gnu-a/9.2-2019.12/binrel/gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu.tar.xz
|
| 71 |
-
tar -xf cpp/toolchains/gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu.tar.xz \
|
| 72 |
-
-C cpp/toolchains
|
| 73 |
-
```
|
| 74 |
-
|
| 75 |
-
BSP SDK 下载地址:
|
| 76 |
-
|
| 77 |
-
| 平台 | BSP SDK |
|
| 78 |
-
| --- | --- |
|
| 79 |
-
| AX650 | <https://github.com/ZHEQIUSHUI/assets/releases/download/ax_3.6.2/msp_50_3.10.2.zip> |
|
| 80 |
-
| AX630C | <https://github.com/ZHEQIUSHUI/assets/releases/download/ax_3.6.2/msp_20e_3.0.0.zip> |
|
| 81 |
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
```bash
|
| 85 |
-
bash cpp/download_bsp.sh all
|
| 86 |
-
```
|
| 87 |
|
| 88 |
-
|
| 89 |
|
| 90 |
```bash
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
编译结果:
|
| 98 |
-
|
| 99 |
-
```text
|
| 100 |
-
cpp/bin/openwakeword_ax650
|
| 101 |
-
cpp/bin/openwakeword_ax630c
|
| 102 |
```
|
| 103 |
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
AX650:
|
| 107 |
|
| 108 |
```bash
|
| 109 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
```
|
| 111 |
|
| 112 |
-
|
| 113 |
|
| 114 |
```bash
|
| 115 |
-
THRESHOLD=0.5 bash cpp/
|
|
|
|
| 116 |
```
|
| 117 |
|
| 118 |
结果分别写入 `outputs/cpp_ax650_batch/` 和 `outputs/cpp_ax630c_batch/`。
|
| 119 |
-
|
| 120 |
详细 C++ 说明见 `cpp/README.md`。
|
| 121 |
|
| 122 |
## 参考
|
|
@@ -133,4 +122,3 @@ THRESHOLD=0.5 bash cpp/run_batch_openwakeword_ax630c.sh
|
|
| 133 |
(峰值窗口实测 0.005 / 0.79);本版本改用包含 TTS 唤醒词正样本的校准集
|
| 134 |
(高响应窗口采样),峰值窗口得分恢复到 0.97 / 1.00(alexa/mycroft/timer 保持 1.0)。
|
| 135 |
校准与编译配方见 https://github.com/AXERA-TECH/openWakeWord.AXERA/pull/1 。
|
| 136 |
-
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
pipeline_tag: audio-classification
|
| 6 |
+
tags:
|
| 7 |
+
- keyword-spotting
|
| 8 |
+
- wake-word
|
| 9 |
+
- axera
|
| 10 |
+
- ax650
|
| 11 |
+
- ax630c
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
# openWakeWord.AXERA
|
| 15 |
|
| 16 |
[openWakeWord](https://github.com/dscripka/openWakeWord) 语音唤醒 AXERA 推理 demo。
|
|
|
|
| 18 |
## 功能
|
| 19 |
|
| 20 |
- 支持 AX650 和 AX630C;
|
| 21 |
+
- 支持 Python SDK 和 C++ 预编译二进制;
|
| 22 |
|
| 23 |
## 目录结构
|
| 24 |
|
|
|
|
| 29 |
│ └── 630C/ # AX630C axmodel
|
| 30 |
├── config/ # STFT/mel 权重
|
| 31 |
├── audio/openwakeword/
|
| 32 |
+
├── scripts/ # Python SDK
|
| 33 |
+
├── cpp/
|
| 34 |
+
│ ├── bin/ # C++ 预编译二进制(ax650 / ax630c)
|
| 35 |
+
│ └── run_*.sh # 板端运行脚本
|
| 36 |
└── README.md
|
| 37 |
```
|
| 38 |
|
| 39 |
+
C++ 源码、构建脚本与 BSP 下载脚本在 GitHub 仓库:
|
| 40 |
+
<https://github.com/AXERA-TECH/openWakeWord.AXERA>
|
| 41 |
+
(HuggingFace 仅发布预编译产物。)
|
| 42 |
|
| 43 |
## Python
|
| 44 |
|
| 45 |
```bash
|
| 46 |
+
# 创建虚拟环境
|
| 47 |
conda create -n OpenWakeWord python=3.10
|
| 48 |
conda activate OpenWakeWord
|
| 49 |
|
| 50 |
+
# 安装 pyaxengine(https://github.com/AXERA-TECH/pyaxengine/releases/latest)
|
| 51 |
pip install axengine-x.x.x-py3-none-any.whl
|
| 52 |
|
| 53 |
pip install -r requirements.txt
|
|
|
|
| 74 |
--output outputs/openwakeword_ax630c.json
|
| 75 |
```
|
| 76 |
|
| 77 |
+
## C++(预编译二进制)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
|
| 79 |
+
板端直接使用 `cpp/bin/` 下的预编译二进制,无需交叉编译。
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
|
| 81 |
+
AX650:
|
| 82 |
|
| 83 |
```bash
|
| 84 |
+
LD_LIBRARY_PATH=<BSP_LIB_DIR> cpp/bin/openwakeword_ax650 \
|
| 85 |
+
--models-dir models/650 \
|
| 86 |
+
--mel-weights config/openwakeword_mel_weights.bin \
|
| 87 |
+
--audio audio/openwakeword/alexa_test.wav \
|
| 88 |
+
--threshold 0.5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
```
|
| 90 |
|
| 91 |
+
AX630C:
|
|
|
|
|
|
|
| 92 |
|
| 93 |
```bash
|
| 94 |
+
LD_LIBRARY_PATH=<BSP_LIB_DIR> cpp/bin/openwakeword_ax630c \
|
| 95 |
+
--models-dir models/630C \
|
| 96 |
+
--mel-weights config/openwakeword_mel_weights.bin \
|
| 97 |
+
--audio audio/openwakeword/alexa_test.wav \
|
| 98 |
+
--threshold 0.5
|
| 99 |
```
|
| 100 |
|
| 101 |
+
也可以使用仓库内的批量运行脚本:
|
| 102 |
|
| 103 |
```bash
|
| 104 |
+
THRESHOLD=0.5 bash cpp/run_batch_openwakeword_ax650.sh # AX650
|
| 105 |
+
THRESHOLD=0.5 bash cpp/run_batch_openwakeword_ax630c.sh # AX630C
|
| 106 |
```
|
| 107 |
|
| 108 |
结果分别写入 `outputs/cpp_ax650_batch/` 和 `outputs/cpp_ax630c_batch/`。
|
|
|
|
| 109 |
详细 C++ 说明见 `cpp/README.md`。
|
| 110 |
|
| 111 |
## 参考
|
|
|
|
| 122 |
(峰值窗口实测 0.005 / 0.79);本版本改用包含 TTS 唤醒词正样本的校准集
|
| 123 |
(高响应窗口采样),峰值窗口得分恢复到 0.97 / 1.00(alexa/mycroft/timer 保持 1.0)。
|
| 124 |
校准与编译配方见 https://github.com/AXERA-TECH/openWakeWord.AXERA/pull/1 。
|
|
|