style: 规范代码风格
This commit is contained in:
nl
parent
3fd36cbf13
commit
50ae952cd6
|
|
@ -18,4 +18,9 @@
|
|||
# along with ASRT. If not, see <https://www.gnu.org/licenses/>.
|
||||
# ============================================================================
|
||||
|
||||
'''
|
||||
@author: nl8590687
|
||||
ASRT语音识别声学特征模块
|
||||
'''
|
||||
|
||||
from .speech_features import *
|
||||
|
|
@ -20,6 +20,12 @@
|
|||
|
||||
# calculate filterbank features. Provides e.g. fbank and mfcc features for use in ASR applications
|
||||
# Author: James Lyons 2012
|
||||
|
||||
'''
|
||||
@author: nl8590687
|
||||
ASRT语音识别声学特征基础库模块，一些基础函数实现
|
||||
'''
|
||||
|
||||
from __future__ import division
|
||||
import numpy
|
||||
from scipy.fftpack import dct
|
||||
|
|
|
|||
|
|
@ -20,10 +20,16 @@
|
|||
|
||||
# This file includes routines for basic signal processing including framing and computing power spectra.
|
||||
# Author: James Lyons 2012
|
||||
|
||||
'''
|
||||
@author: nl8590687
|
||||
ASRT语音识别声学特征计算的信号处理计算的函数库
|
||||
'''
|
||||
|
||||
import decimal
|
||||
import numpy
|
||||
import math
|
||||
import logging
|
||||
import math
|
||||
import numpy
|
||||
|
||||
|
||||
def round_half_up(number):
|
||||
|
|
@ -135,7 +141,7 @@ def logpowspec(frames, NFFT, norm=1):
|
|||
:param norm: If norm=1, the log power spectrum is normalised so that the max value (across all frames) is 0.
|
||||
:returns: If frames is an NxD matrix, output will be Nx(NFFT/2+1). Each row will be the log power spectrum of the corresponding frame.
|
||||
"""
|
||||
ps = powspec(frames, NFFT);
|
||||
ps = powspec(frames, NFFT)
|
||||
ps[ps <= 1e-30] = 1e-30
|
||||
lps = 10 * numpy.log10(ps)
|
||||
if norm:
|
||||
|
|
|
|||
|
|
@ -20,13 +20,11 @@
|
|||
|
||||
"""
|
||||
@author: nl8590687
|
||||
ASRT语音识别内置声学特征提取模块
|
||||
|
||||
ASRT语音识别内置声学特征提取模块，定义了几个常用的声学特征类
|
||||
"""
|
||||
|
||||
import random
|
||||
import numpy as np
|
||||
import math
|
||||
from scipy.fftpack import fft
|
||||
from .base import mfcc, delta, logfbank
|
||||
|
||||
|
|
@ -38,10 +36,12 @@ class SpeechFeatureMeta():
|
|||
'''
|
||||
def __init__(self, framesamplerate = 16000):
|
||||
self.framesamplerate = framesamplerate
|
||||
pass
|
||||
|
||||
def run(self, wavsignal, fs = 16000):
|
||||
raise NotImplementedError('[ASRT] Get speech feature function is not implemented. Please define "a run method"')
|
||||
'''
|
||||
run method
|
||||
'''
|
||||
raise NotImplementedError('[ASRT] `run()` method is not implemented.')
|
||||
|
||||
class MFCC(SpeechFeatureMeta):
|
||||
'''
|
||||
|
|
@ -56,7 +56,12 @@ class MFCC(SpeechFeatureMeta):
|
|||
:param nfilt: the number of filters in the filterbank, default 26.
|
||||
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
|
||||
'''
|
||||
def __init__(self, framesamplerate = 16000, winlen=0.025, winstep=0.01, numcep=13, nfilt=26, preemph=0.97):
|
||||
def __init__(self, framesamplerate = 16000,
|
||||
winlen=0.025,
|
||||
winstep=0.01,
|
||||
numcep=13,
|
||||
nfilt=26,
|
||||
preemph=0.97):
|
||||
self.framesamplerate = framesamplerate
|
||||
self.winlen = winlen
|
||||
self.winstep = winstep
|
||||
|
|
@ -73,7 +78,8 @@ class MFCC(SpeechFeatureMeta):
|
|||
'''
|
||||
wavsignal = np.array(wavsignal, dtype=np.float)
|
||||
# 获取输入特征
|
||||
feat_mfcc=mfcc(wavsignal[0], samplerate=self.framesamplerate, winlen=self.winlen, winstep=self.winstep, numcep=self.numcep, nfilt=self.nfilt, preemph=self.preemph)
|
||||
feat_mfcc=mfcc(wavsignal[0], samplerate=self.framesamplerate, winlen=self.winlen,
|
||||
winstep=self.winstep, numcep=self.numcep, nfilt=self.nfilt, preemph=self.preemph)
|
||||
feat_mfcc_d=delta(feat_mfcc, 2)
|
||||
feat_mfcc_dd=delta(feat_mfcc_d, 2)
|
||||
# 返回值分别是mfcc特征向量的矩阵及其一阶差分和二阶差分矩阵
|
||||
|
|
@ -94,9 +100,13 @@ class Logfbank(SpeechFeatureMeta):
|
|||
return wav_feature
|
||||
|
||||
class Spectrogram(SpeechFeatureMeta):
|
||||
'''
|
||||
ASRT语音识别内置的语谱图声学特征提取类
|
||||
'''
|
||||
def __init__(self, framesamplerate = 16000, timewindow = 25, timeshift = 10):
|
||||
self.time_window = timewindow
|
||||
self.window_length = int(framesamplerate / 1000 * self.time_window) # 计算窗长度的公式，目前全部为400固定值
|
||||
self.timeshift = timeshift
|
||||
|
||||
'''
|
||||
# 保留将来用于不同采样频率
|
||||
|
|
@ -109,7 +119,7 @@ class Spectrogram(SpeechFeatureMeta):
|
|||
super().__init__(framesamplerate)
|
||||
|
||||
def run(self, wavsignal, fs = 16000):
|
||||
if(16000 != fs):
|
||||
if fs != 16000:
|
||||
raise ValueError('[Error] ASRT currently only supports wav audio files with a sampling rate of 16000 Hz, but this audio is ' + str(fs) + ' Hz. ')
|
||||
|
||||
# wav波形 加时间窗以及时移10ms
|
||||
|
|
@ -118,7 +128,7 @@ class Spectrogram(SpeechFeatureMeta):
|
|||
|
||||
wav_arr = np.array(wavsignal)
|
||||
#wav_length = len(wavsignal[0])
|
||||
wav_length = wav_arr.shape[1]
|
||||
#wav_length = wav_arr.shape[1]
|
||||
|
||||
range0_end = int(len(wavsignal[0])/fs*1000 - time_window) // 10 + 1 # 计算循环终止的位置，也就是最终生成的窗数
|
||||
data_input = np.zeros((range0_end, window_length // 2), dtype = np.float) # 用于存放最终的频率特征数据
|
||||
|
|
@ -129,10 +139,7 @@ class Spectrogram(SpeechFeatureMeta):
|
|||
p_end = p_start + 400
|
||||
|
||||
data_line = wav_arr[0, p_start:p_end]
|
||||
|
||||
data_line = data_line * self.w # 加窗
|
||||
|
||||
#data_line = np.abs(fft(data_line)) / wav_length
|
||||
data_line = np.abs(fft(data_line))
|
||||
|
||||
data_input[i]=data_line[0: window_length // 2] # 设置为400除以2的值（即200）是取一半数据，因为是对称的
|
||||
|
|
@ -143,11 +150,12 @@ class Spectrogram(SpeechFeatureMeta):
|
|||
|
||||
class SpecAugment(SpeechFeatureMeta):
|
||||
'''
|
||||
复现谷歌SpecAugment数据增强算法
|
||||
复现谷歌SpecAugment数据增强特征算法，基于Spectrogram语谱图基础特征
|
||||
'''
|
||||
def __init__(self, framesamplerate = 16000, timewindow = 25, timeshift = 10):
|
||||
self.time_window = timewindow
|
||||
self.window_length = int(framesamplerate / 1000 * self.time_window) # 计算窗长度的公式，目前全部为400固定值
|
||||
self.timeshift = timeshift
|
||||
|
||||
'''
|
||||
# 保留将来用于不同采样频率
|
||||
|
|
@ -160,7 +168,7 @@ class SpecAugment(SpeechFeatureMeta):
|
|||
super().__init__(framesamplerate)
|
||||
|
||||
def run(self, wavsignal, fs = 16000):
|
||||
if(16000 != fs):
|
||||
if fs != 16000:
|
||||
raise ValueError('[Error] ASRT currently only supports wav audio files with a sampling rate of 16000 Hz, but this audio is ' + str(fs) + ' Hz. ')
|
||||
|
||||
# wav波形 加时间窗以及时移10ms
|
||||
|
|
@ -169,7 +177,7 @@ class SpecAugment(SpeechFeatureMeta):
|
|||
|
||||
wav_arr = np.array(wavsignal)
|
||||
#wav_length = len(wavsignal[0])
|
||||
wav_length = wav_arr.shape[1]
|
||||
#wav_length = wav_arr.shape[1]
|
||||
|
||||
range0_end = int(len(wavsignal[0])/fs*1000 - time_window) // 10 + 1 # 计算循环终止的位置，也就是最终生成的窗数
|
||||
data_input = np.zeros((range0_end, window_length // 2), dtype = np.float) # 用于存放最终的频率特征数据
|
||||
|
|
@ -180,10 +188,7 @@ class SpecAugment(SpeechFeatureMeta):
|
|||
p_end = p_start + 400
|
||||
|
||||
data_line = wav_arr[0, p_start:p_end]
|
||||
|
||||
data_line = data_line * self.w # 加窗
|
||||
|
||||
#data_line = np.abs(fft(data_line)) / wav_length
|
||||
data_line = np.abs(fft(data_line))
|
||||
|
||||
data_input[i]=data_line[0: window_length // 2] # 设置为400除以2的值（即200）是取一半数据，因为是对称的
|
||||
|
|
@ -199,17 +204,13 @@ class SpecAugment(SpeechFeatureMeta):
|
|||
v_start = random.randint(1,data_input.shape[1])
|
||||
v_width = random.randint(1,100)
|
||||
|
||||
if(mode <= 60): # 正常特征 60%
|
||||
if mode <= 60: # 正常特征 60%
|
||||
pass
|
||||
elif(mode > 60 and mode <=75): # 横向遮盖 15%
|
||||
elif 60 < mode <=75: # 横向遮盖 15%
|
||||
data_input[h_start:h_start+h_width,:] = 0
|
||||
pass
|
||||
elif(mode > 75 and mode <= 90): # 纵向遮盖 15%
|
||||
elif 75 < mode <= 90: # 纵向遮盖 15%
|
||||
data_input[:,v_start:v_start+v_width] = 0
|
||||
pass
|
||||
else: # 两种遮盖叠加 10%
|
||||
data_input[h_start:h_start+h_width,:v_start:v_start+v_width] = 0
|
||||
pass
|
||||
|
||||
|
||||
return data_input
|
||||
|
|
|
|||
Loading…
Reference in New Issue