Speaker Diarization Community-1 Ascend NPU 部署指南

/data/ysws/agentsp/speaker-diarization-community-1-ascend/
├── inference.py          # 精度测试脚本
├── log.txt               # 测试日志
├── README.md             # 本文档
├── test_audio_1s.pt     # 测试音频 (1秒)
└── test_audio_3s.pt     # 测试音频 (3秒)

docker exec -it test-modelagent bash

source /usr/local/Ascend/ascend-toolkit/set_env.sh

cd /data/ysws/agentsp/speaker-diarization-community-1-ascend/
python3 inference.py

import torch
from pyannote.audio import Pipeline

pipeline = Pipeline.from_pretrained("/data/ysws/agentsp/speaker-diarization-community-1")
pipeline.to("npu:0")

waveform = torch.randn(1, 16000)
audio = {"waveform": waveform, "sample_rate": 16000}

output = pipeline(audio)
for turn, speaker in output.speaker_diarization:
    print(f"{speaker} speaks between t={turn.start:.3f}s and t={turn.end:.3f}s")

import torch
from pyannote.audio import Model

seg_model = Model.from_pretrained("/data/ysws/agentsp/speaker-diarization-community-1/segmentation")
seg_model.to("npu:0")

waveform = torch.randn(1, 16000)
output = seg_model(waveform.to("npu:0"))
print(f"Output shape: {output.shape}")  # (batch, frames, 7)

import torch
from pyannote.audio import Model

emb_model = Model.from_pretrained("/data/ysws/agentsp/speaker-diarization-community-1/embedding")
emb_model.to("npu:0")

waveform = torch.randn(1, 16000)
embedding = emb_model(waveform.to("npu:0"))
print(f"Embedding shape: {embedding.shape}")  # (batch, 256)