语音增强,使用掩蔽方法和映射方法对应论文:A Two-stage Speech Enhancement Method Based on Deep Learning Network所使用的平台、库及版本:
PyCharm 2021.2.3
matlab R2019a
python 3.7.11
keras 2.2.0
tensorflow 1.14.0
librosa 0.8.1
各函数介绍:
generate_data.py 生成训练数据函数
SpectralSubtraction.py 谱减法函数
IBM.py 训练IBM方法函数
Mapping.py 训练Mapping方法函数
mian.py 提出方法的主函数:计算了两种方法(IBM,Bss_Mapping)对含有两种噪声(白噪声和机关枪噪声)在6种信噪比(-5,0,5,10,15,20)下的语音进行增强的结果,输出个种情况的增强语音
measure_pesq.m 评价语音失真函数:计算增强语音的PESQ STIO指标
measure_snr.m 评价信噪比函数:计算增强语音的SNR,SegSNR,FWSegSNR指标
运行步骤:
1.运行mian.py
2.运行measure_pesq.m
3.运行measure_snr.m
得到数据后在Excel表格中画图即可得到论文中的图表
.
└── AIACT2022
├── DC_block.m
├── FFTNXCorr.m
├── IBM_train.py
├── Mapping_train.py
├── SpectralSubtraction.py
├── __pycache__
│ ├── SpectralSubtraction.cpython-36.pyc
│ └── basicfunctions.cpython-36.pyc
├── apply_VAD.m
├── apply_filter.m
├── apply_filters.m
├── basicfunctions.py
├── clean.wav
├── comp_fwseg.m
├── comp_snr.m
├── crude_align.m
├── fix_power_level.m
├── generate_data.py
├── id_searchwindows.m
├── id_utterances.m
├── input_filter.m
├── main.py
├── measure_pesq.m
├── measure_snr.m
├── measure_test.m
├── pesq.m
├── pesq_n.m
├── pesq_psychoacoustic_model.m
├── pow_of.m
├── setup_global.m
├── split_align.m
├── stoi.m
├── testfunctions.py
├── time_align.m
├── timit_clean_selected_train.wav
├── utterance_locate.m
├── utterance_split.m
├── 代码说明.txt
└── 数据分析.xlsx
2 directories, 38 files
评论