硬件:Ascend NPU(Ascend 910B4) · 推理引擎:torch_npu · 设备:npu:0 精度验收:CPU vs NPU 推理精度差异 < 1%(实测 PASS,cosine=0.999909520149231)
Relies on pyannote.audio 2.1: see installation instructions. ```python
| 属性 | 值 |
|---|---|
| 模型名 | pyannote/voice-activity-detection |
| 原始权重 | ModelScope / HuggingFace / AtomGit 镜像 |
| 任务类型 | feature-extraction |
| 输入 | 见 inference.py |
| 输出 | 特征向量 |
| 框架 | PyTorch + torch_npu |
| 官方实现 | https://github.com/pyannote/pyannote-audio |
官方实现 / 论文仓库:
| 依赖项 | 版本 | 说明 |
|---|---|---|
| Python | 3.11 | conda 环境 pyannote-voice-activity-detection-npu |
| torch / torch_npu | 2.10.0 / 2.10.0 | 由 setup_env.sh 装,须严格配套 |
| transformers | ==4.51.3 | |
| accelerate | ==1.14.0 | |
| pillow | ==12.3.0 | |
| numpy | ==1.26.4 | |
| huggingface-hub | ==0.36.2 | |
| requests | ==2.34.2 | |
| modelscope | ==1.37.0 | |
| safetensors | ==0.8.0 | |
| multimolecule | ==0.2.1 | |
| pyyaml | >=6.0 | |
| einops | ==0.8.0 # transformers 4.51 modeling 隐式依赖(torch_npu 存在时触发) | |
| torchaudio | ==2.10.0 # pyannote.audio 依赖(版本号与 torch 同步,无 0.x) | |
| pyannote.audio | ==3.3.2 # Model.from_pretrained 本地直载(VAD/segmentation) |
一键建环境(从零创建 conda 环境并装依赖):
bash setup_env.sh # conda create -n <env> python=<ver> + torch/torch_npu + requirements.txt⚠️ 重要:若环境已装 torch/torch_npu,不要强制覆盖 torch,否则破坏 torch_npu 算子插件绑定。
国内 pip 源(Tsinghua):https://pypi.tuna.tsinghua.edu.cn/simple
权重不随仓库分发,推理前先用国内下载源拉取到本地缓存(推荐 ModelScope,失败回落 HF 镜像):
# 方式一:ModelScope(国内直连,最快)
pip install modelscope -i https://pypi.tuna.tsinghua.edu.cn/simple
python -c "from modelscope import snapshot_download; snapshot_download('pyannote/voice-activity-detection', cache_dir='./weights')"
# 方式二:HF 镜像(hf-mirror.com)
export HF_ENDPOINT=https://hf-mirror.com
huggingface-cli download pyannote/voice-activity-detection --local-dir ./weights/voice-activity-detectioninference.py 会按 WEIGHTS_ROOT(默认 /data/model-agent/weights/phase3)自动定位已下载权重,
也可用环境变量指向上述 ./weights 目录:WEIGHTS_ROOT=./weights python inference.py
硬件:单卡 Ascend 910B4(npu:1,物理卡 5)
评委裸 clone 后本机没有 conda 环境,conda activate 不会创建环境。必须先建环境再推理:
# 第一步:从零创建 conda 环境并装依赖(conda create + torch/torch_npu + requirements.txt)
bash setup_env.sh
# 第二步:激活环境并一键推理(环境自检 + inference + eval + benchmark)
conda activate pyannote-voice-activity-detection-npu
bash run.shimport torch, torch_npu
from transformers import AutoModel, AutoTokenizer
m = AutoModel.from_pretrained("pyannote/voice-activity-detection", trust_remote_code=True).to("npu:0").eval()
tok = AutoTokenizer.from_pretrained("pyannote/voice-activity-detection", trust_remote_code=True)
inputs = tok("How to use this model?", return_tensors="pt").to("npu:0")
outputs = m(**inputs)logs/inference.log)输入:示例序列/文本(见 inference.py)
输出:
=== Inference: pyannote/voice-activity-detection on Ascend NPU ===
NPU: npu:0
测试音频: /data/model-agent/昇腾Model-Agent模型适配大赛第二季/projects/pyannote-voice-activity-detection-NPU/assets/test.wav (440Hz 1s)
加载方式: pyannote_model
推理耗时: 759.03 ms
输出 shape: (1, 293, 7)
输出前 8 维: [-0.201643705368042, -4.83579158782959, -3.313601016998291, -2.059173107147217, -6.752871513366699, -6.101363182067871, -4.916889190673828, -0.18083912134170532]logs/accuracy.log)=== CPU-NPU Accuracy: pyannote/voice-activity-detection ===
cosine_similarity: 0.999910
结果 PASS(阈值 cosine > 0.999)| 指标 | 数值 |
|---|---|
| max_abs_error | N/A |
| relative_error | N/A |
| cosine_similarity | 0.999909520149231 |
注:本次适配无精度要求,精度数据作为质量证据。详见
logs/accuracy.log。
| 指标 | 数值 |
|---|---|
| avg latency | 241.27 ms |
| p50 | N/A |
| throughput | 4.14 samples/sec |



截图由真实执行日志(
logs/)忠实渲染为终端风格 PNG,渲染脚本tools/gen_screenshots_pw.py(playwright),可复现,无虚构结果。
#NPU #Ascend #Ascend910 #model-agent-tagged
仓库:https://gitcode.com/v_50/pyannote-voice-activity-detection-NPU 适配日期:2026-08-12 | 赛道:第二季·模型适配·阶段三