PANN 推断

panns_inference为音频标记和声音事件检测提供了一个易于使用的 Python 接口。音频标记和声音事件检测模型从 PANN 训练：用于音频模式识别的大规模预训练音频神经网络：https ://github.com/qiuqiangkong/audioset_tagging_cnn

安装

PyTorch>=1.0 是必需的。

$ pip install panns-inference

用法

$ python3 example.py

例如：

import librosa
import panns_inference
from panns_inference import AudioTagging, SoundEventDetection, labels

audio_path = 'examples/R9_ZSCveAHg_7s.wav'
(audio, _) = librosa.core.load(audio_path, sr=32000, mono=True)
audio = audio[None, :]  # (batch_size, segment_samples)

print('------ Audio tagging ------')
at = AudioTagging(checkpoint_path=None, device='cuda')
(clipwise_output, embedding) = at.inference(audio)

print('------ Sound event detection ------')
sed = SoundEventDetection(checkpoint_path=None, device='cuda')
framewise_output = sed.inference(audio)

结果

------ Audio tagging ------
Checkpoint path: /root/panns_data/Cnn14_mAP=0.431.pth
GPU number: 1
Speech: 0.893
Telephone bell ringing: 0.754
Inside, small room: 0.235
Telephone: 0.183
Music: 0.092
Ringtone: 0.047
Inside, large room or hall: 0.028
Alarm: 0.014
Animal: 0.009
Vehicle: 0.008
------ Sound event detection ------
Checkpoint path: /root/panns_data/Cnn14_mAP=0.431.pth
GPU number: 1
Save fig to results/sed_result.pdf

声音事件检测图：

引用

[1] Kong, Qiuqiang, Yin Cao, Turab Iqbal, Yuxuan Wang, Wenwu Wang, and Mark D. Plumbley。“PANN：用于音频模式识别的大规模预训练音频神经网络。” arXiv 预印本 arXiv:1912.10211 (2019)。