产品上手指引
Linux PC
AI 加速卡
LLM-8850 Card
大语言模型
实时 AI 语音助手
火山引擎语音助手
离线语音识别
工业控制
Ezdata
Ethernet 摄像头
PoECAM
Wi-Fi 摄像头
Unit CamS3/-5MP
AI 摄像头
LoRa & LoRaWAN
电机驱动
恢复出厂固件教程
CosyVoice2
- 手动下载模型 并上传到 raspberrypi5，或者通过以下命令拉取模型仓库。
提示
如果没有安装 git lfs，先参考git lfs 安装说明进行安装。
git clone https://huggingface.co/AXERA-TECH/CosyVoice2文件说明：
m5stack@raspberrypi:~/rsp/CosyVoice2 $ ls -lh
total 26M
drwxrwxr-x 2 m5stack m5stack 4.0K Sep 27 15:54 asset
-rwxrwxr-x 1 m5stack m5stack 0 Sep 18 14:32 config.json
drwxrwxr-x 2 m5stack m5stack 4.0K Sep 5 19:12 CosyVoice-BlankEN-Ax650-prefill_512
drwxrwxr-x 2 m5stack m5stack 4.0K Sep 5 19:11 frontend-onnx
-rwxrwxr-x 1 m5stack m5stack 9.3M Sep 27 15:33 main_api_ax650
-rwxrwxr-x 1 m5stack m5stack 1.9M Sep 27 15:33 main_api_axcl_aarch64
-rwxrwxr-x 1 m5stack m5stack 2.0M Sep 27 15:33 main_api_axcl_x86
-rwxrwxr-x 1 m5stack m5stack 9.2M Sep 18 09:35 main_ax650
-rwxrwxr-x 1 m5stack m5stack 1.8M Sep 27 15:33 main_axcl_aarch64
-rwxrwxr-x 1 m5stack m5stack 1.9M Sep 27 15:33 main_axcl_x86
drwxrwxr-x 3 m5stack m5stack 4.0K Sep 27 15:33 onnxruntime-linux-aarch64-1.23.0
drwxrwxr-x 3 m5stack m5stack 4.0K Sep 27 15:33 onnxruntime-linux-x64-1.23.0
drwxrwxr-x 2 m5stack m5stack 4.0K Sep 15 10:07 prompt_files
-rwxrwxr-x 1 m5stack m5stack 7.9K Sep 27 15:33 README.md
-rwxrwxr-x 1 m5stack m5stack 895 Sep 27 15:33 run_api_ax650.sh
-rwxrwxr-x 1 m5stack m5stack 997 Sep 27 15:33 run_api_axcl_aarch64.sh
-rwxrwxr-x 1 m5stack m5stack 989 Sep 27 15:33 run_api_axcl_x86.sh
-rwxrwxr-x 1 m5stack m5stack 865 Sep 27 15:33 run_ax650.sh
-rwxrwxr-x 1 m5stack m5stack 967 Sep 27 15:33 run_axcl_aarch64.sh
-rwxrwxr-x 1 m5stack m5stack 959 Sep 27 15:33 run_axcl_x86.sh
drwxrwxr-x 5 m5stack m5stack 4.0K Sep 27 15:33 scripts
drwxrwxr-x 2 m5stack m5stack 4.0K Sep 27 15:33 token2wav-axmodels- 创建虚拟环境
python -m venv cosyvoice- 激活虚拟环境
source cosyvoice/bin/activate- 安装依赖包
pip install -r scripts/requirements.txt- 启动 tokenizer 解析器
cd scripts
python cosyvoice2_tokenizer.py- 运行 tokenizer 服务，Host ip 默认为 localhost，端口号设置为 12345，运行后信息如下：
(cosyvoice) m5stack@raspberrypi:~/rsp/CosyVoice2/scripts $ python cosyvoice2_tokenizer.py
[14990, 1879]
http://localhost:12345提示
以下操作需要新建一个 raspberrypi 的终端。
- 使用 CosyVoice2 模型单次推理，输出音频文件为 output.wav
./run_axcl_aarch64.sh成功启动后信息如下：
m5stack@raspberrypi:~/rsp/CosyVoice2 $ ./run_axcl_aarch64.sh
rm: cannot remove 'output*.wav': No such file or directory
[I][ main][ 291]: device: 0
[I][ run][ 30]: AXCLWorker start with devid 0
[I][ Init][ 135]: LLM init start
[I][ Init][ 34]: connect http://127.0.0.1:12345 ok
bos_id: 0, eos_id: 1773
3% | ██ | 1 / 27 [0.00s<0.03s, 1000.00 count 7% | ███ | 2 / 27 [0.13s<1.81s, 14.93 count/s] embed_selector init ok[I][ Init][ 174]: attr.axmodel_num:24
11% | ████ | 3 / 27 [1.35s<12.13s, 2.23 count/s 14% | ███████████████████████████████ | 26 / 27 [7.92s<8.22s, 3.28 count/s]100% | ████████████████████████████████ | 27 / 27 [10.59s<10.59s, 2.55 count/s] init post axmodel ok,remain_cmm(6323 MB)
[I][ Init][ 225]: max_token_len : 1023
[I][ Init][ 230]: kv_cache_size : 128, kv_cache_num: 1023
[I][ Init][ 238]: prefill_token_num : 128
[I][ Init][ 242]: grp: 1, prefill_max_token_num : 1
[I][ Init][ 242]: grp: 2, prefill_max_token_num : 128
[I][ Init][ 242]: grp: 3, prefill_max_token_num : 256
[I][ Init][ 242]: grp: 4, prefill_max_token_num : 384
[I][ Init][ 242]: grp: 5, prefill_max_token_num : 512
[I][ Init][ 246]: prefill_max_token_num : 512
________________________
| ID| remain cmm(MB)|
========================
| 0| 6323|
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
[I][ Init][ 308]: LLM init ok
2025-09-29 11:09:20.814330123 [W:onnxruntime:, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"
inputs:
mel: 1 x 80 x 50
output:
s: 1 x 1 x 24000
inputs:
mel: 1 x 80 x 58
output:
s: 1 x 1 x 27840
[I][ Init][ 213]: Token2Wav init ok, remain_cmm(5863 MB)
[I][ Run][ 449]: input token num : 142, prefill_split_num : 2
[I][ Run][ 483]: input_num_token:128
[I][ Run][ 483]: input_num_token:14
[I][ Run][ 642]: ttft: 339.72 ms
[Main/Token2Wav Thread] Processing batch of 28 tokens...
Successfully saved audio to output_0.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 53 tokens...
Successfully saved audio to output_1.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_2.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_3.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_4.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_5.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_6.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_7.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_8.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_9.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_10.wav (32-bit Float PCM).
[I][ Run][ 765]: hit eos, llm finished
[I][ Run][ 795]: llm finished
[Main/Token2Wav Thread] Buffer is empty and LLM finished. Exiting.
[I][ Run][ 800]: total decode tokens:300
[N][ Run][ 801]: hit eos, decode avg 14.54 token/s
Successfully saved audio to output_11.wav (32-bit Float PCM).
Successfully saved audio to output.wav (32-bit Float PCM).
[I][ tts][ 225]: tts total use time: 21.555 s
Voice generation pipeline completed.
[I][ run][ 80]: AXCLWorker exit with devid 0- 修改 ‘run_axcl_aarch64.sh’ 中 continue 为 1 ，再次执行，进入 CosyVoice2 交互推理模式，输出音频文件为 output.wav
m5stack@raspberrypi:~/rsp/CosyVoice2 $ cat ./run_axcl_aarch64.sh
export LD_LIBRARY_PATH=onnxruntime-linux-aarch64-1.23.0/lib:$LD_LIBRARY_PATH
LLM_DIR=CosyVoice-BlankEN-Ax650-prefill_512/
TOKEN2WAV_DIR=token2wav-axmodels/
rm output*.wav
./main_axcl_aarch64 \
--template_filename_axmodel "${LLM_DIR}/qwen2_p128_l%d_together.axmodel" \
--token2wav_axmodel_dir $TOKEN2WAV_DIR \
--n_timesteps 10 \
--axmodel_num 24 \
--bos 0 --eos 0 \
--filename_tokenizer_model "http://127.0.0.1:12345" \
--filename_post_axmodel "${LLM_DIR}/qwen2_post.axmodel" \
--filename_decoder_axmodel "${LLM_DIR}/llm_decoder.axmodel" \
--filename_tokens_embed "${LLM_DIR}/model.embed_tokens.weight.bfloat16.bin" \
--filename_llm_embed "${LLM_DIR}/llm.llm_embedding.float16.bin" \
--filename_speech_embed "${LLM_DIR}/llm.speech_embedding.float16.bin" \
--continue 1 \
--devices "0," \
--prompt_files prompt_files \
--text "君不见黄河之水天上来，奔流到海不复回。君不见高堂明镜悲白发，朝如青丝暮成雪。"
chmod 777 output*.wav成功启动后信息如下：
m5stack@raspberrypi:~/rsp/CosyVoice2 $ ./run_axcl_aarch64.sh
[I][ main][ 291]: device: 0
[I][ run][ 30]: AXCLWorker start with devid 0
[I][ Init][ 135]: LLM init start
[I][ Init][ 34]: connect http://127.0.0.1:12345 ok
bos_id: 0, eos_id: 1773
3% | ██ | 1 / 27 [0.00s<0.03s, 1000.00 count 7% | ████████████████████████████████ | 27 / 27 [10.57s<10.57s, 2.55 count/s] init post axmodel ok,remain_cmm(6323 MB)
[I][ Init][ 225]: max_token_len : 1023
[I][ Init][ 230]: kv_cache_size : 128, kv_cache_num: 1023
[I][ Init][ 238]: prefill_token_num : 128
[I][ Init][ 242]: grp: 1, prefill_max_token_num : 1
[I][ Init][ 242]: grp: 2, prefill_max_token_num : 128
[I][ Init][ 242]: grp: 3, prefill_max_token_num : 256
[I][ Init][ 242]: grp: 4, prefill_max_token_num : 384
[I][ Init][ 242]: grp: 5, prefill_max_token_num : 512
[I][ Init][ 246]: prefill_max_token_num : 512
________________________
| ID| remain cmm(MB)|
========================
| 0| 6323|
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
[I][ Init][ 308]: LLM init ok
2025-09-29 11:22:34.715409537 [W:onnxruntime:, device_discovery.cc:164 DiscoverDevicesForPlatform] GPU device discovery failed: device_discovery.cc:89 ReadFileContents Failed to open file: "/sys/class/drm/card1/device/vendor"
inputs:
mel: 1 x 80 x 50
output:
s: 1 x 1 x 24000
inputs:
mel: 1 x 80 x 58
output:
s: 1 x 1 x 27840
[I][ Init][ 213]: Token2Wav init ok, remain_cmm(5863 MB)
[I][ Run][ 449]: input token num : 142, prefill_split_num : 2
[I][ Run][ 483]: input_num_token:128
[I][ Run][ 483]: input_num_token:14
[I][ Run][ 642]: ttft: 342.93 ms
[Main/Token2Wav Thread] Processing batch of 28 tokens...
Successfully saved audio to output_0.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 53 tokens...
Successfully saved audio to output_1.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_2.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_3.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_4.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_5.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_6.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_7.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_8.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_9.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
[I][ Run][ 765]: hit eos, llm finished
[I][ Run][ 795]: llm finished
[I][ Run][ 800]: total decode tokens:278
[N][ Run][ 801]: hit eos, decode avg 14.37 token/s
Successfully saved audio to output_10.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Buffer is empty and LLM finished. Exiting.
Successfully saved audio to output_11.wav (32-bit Float PCM).
Successfully saved audio to output.wav (32-bit Float PCM).
[I][ tts][ 225]: tts total use time: 20.305 s
Voice generation pipeline completed.
Type "q" to exit, Ctrl+c to stop current running
text >> 很高兴认识你，Nice to meet you
[I][ Run][ 449]: input token num : 115, prefill_split_num : 1
[I][ Run][ 483]: input_num_token:115
[I][ Run][ 642]: ttft: 153.41 ms
[Main/Token2Wav Thread] Processing batch of 28 tokens...
Successfully saved audio to output_0.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 53 tokens...
Successfully saved audio to output_1.wav (32-bit Float PCM).
[Main/Token2Wav Thread] Processing batch of 78 tokens...
Successfully saved audio to output_2.wav (32-bit Float PCM).
[I][ Run][ 765]: hit eos, llm finished
[I][ Run][ 795]: llm finished
[I][ Run][ 800]: total decode tokens:93
[N][ Run][ 801]: hit eos, decode avg 15.73 token/s
[Main/Token2Wav Thread] Buffer is empty and LLM finished. Exiting.
Successfully saved audio to output_3.wav (32-bit Float PCM).
Successfully saved audio to output.wav (32-bit Float PCM).
[I][ tts][ 225]: tts total use time: 6.617 s
Voice generation pipeline completed.
text >>