DTMF双音多频信号拨号音识别Python实现

DTMF双音多频信号拨号音识别Python实现

本例仅用于记录笔者的一次大作业。代码源于网络大神，进行修改和滤波处理后得到了理想的结果。

import wave
import numpy as np
import matplotlib.pyplot as plt
from scipy import signaldef s_add_wn(x: float, snr: float):        #增加噪音本例未用到Ps = np.sum(abs(x) ** 2) / len(x)Pn = Ps / (10 ** ((snr / 10)))noise = np.random.randn(len(x)) * np.sqrt(Pn)signal_add_noise = x + noisereturn signal_add_noisedef filter_bp(x, fs, wl, wh):fN = 3fc = fs / 2w1c = wl / fcw2c = wh / fcb, a = signal.butter(fN, [w1c, w2c], 'bandpass')x_filter = signal.filtfilt(b, a, x)return x_filterdef dtmf_encoder(number):m = 0.5  # 每个音调的长度s = 0.5  # 音调之间的间隔dtmf_mapping = {'1': (697, 1209),'2': (697, 1336),'3': (697, 1477),'A': (697, 1633),'4': (770, 1209),'5': (770, 1336),'6': (770, 1477),'B': (770, 1633),'7': (852, 1209),'8': (852, 1336),'9': (852, 1477),'C': (852, 1633),'*': (941, 1209),'0': (941, 1336),'#': (941, 1477),'D': (941, 1633), }if len(number) == 1:s = 0x = np.array([])x = np.zeros(int(s * framerate))n = np.arange(0, int(m * framerate))for num in number:p = np.sin(2 * np.pi * (dtmf_mapping[num][0] / framerate) * n) + np.sin(2 * np.pi * (dtmf_mapping[num][1] / framerate) * n)space = np.zeros(int(s * framerate))x = np.concatenate((x, p, space))return xdef wave_write(wave_file, wave_data, framerate, sampwidth=2, channels=1):f = wave.open(wave_file, "wb")# 配置声道数、量化位数和取样频率f.setnchannels(channels)f.setsampwidth(sampwidth)f.setframerate(framerate)# 将wav_data转换为二进制数据写入文件wave_data = wave_data * int(20000 / max(abs(wave_data)))wave_data = np.int16(wave_data)f.writeframes(bytes())f.close()def signal_energy(signal, win=80, framerate=48000):print("signal length:", len(signal))w = np.reshape(signal[:int(len(signal) / win) * win], (-1, win))# print(w)print(w.shape)# this is a 10ms interval..# compute energy of the tonew_energy = np.sum(w * w, axis=1)# plot ittime = np.arange(0, len(w_energy)) * (len(signal) / len(w_energy) / framerate)plt.figure(figsize=(20, 4))plt.plot(time, w_energy)# from the plot, a threshold of 200 is enough to separate sound and silence
# based on that we can define a function that returns the start and stop indices of the tones
def split_dtmf(x, th=5, win=80):edges = []w = np.reshape(x[:int(len(x) / win) * win],(-1, win))  # 假设x是一个包含128个元素的数组，而win为8，那么这段代码会从x中选取16个元素（因为128 // 8 = 16），然后将它们重塑成一个16行（每行8个元素）的二维数组print(w)w_e = np.sum(w * w, axis=1)  # w_e是通过对w的每一行（axis=1）进行平方求和   对每行的80个元素进行平方求和L = len(w_e)print("w_e", w_e)idx = 0while idx < L:while idx < L and w_e[idx] < th:idx = idx + 1if idx >= L:breaki = idxwhile i < L and w_e[i] > th:i = i + 1if i * win - idx * win > 2500:edges.append((idx * win, i * win))print(idx, i)idx = ireturn edgesdef decode_dtmf(x, th=5, edges=None):L_Freqs = np.array([697.0, 770.0, 852.0, 941.0])H_Freqs = np.array([1209.0, 1336.0, 1477.0, 1633.0])KEYS = [['1', '2', '3'], ['4', '5', '6'], ['7', '8', '9'], ['*', '0', '#', 'D']]L_RANGE = (680.0, 960.0)H_RANGE = (1180.0, 1700.0)number = []if edges is None:edges = split_dtmf(x, th)for edge in edges:# compute dft of tone segmentX = np.abs(np.fft.fft(x[edge[0]:edge[1]]))print("edge", edge)N = len(X)print("N", N)res = float(framerate) / N  # 每个DFT仓的分辨率print("res", res)# look for peak in low frequency rangea = int(L_RANGE[0] / res)b = int(L_RANGE[1] / res)lo = a + np.argmax(X[a:b])# look for peak in high frequency rangea = int(H_RANGE[0] / res)b = int(H_RANGE[1] / res)hi = a + np.argmax(X[a:b])row = np.argmin(abs(L_Freqs - lo * res))col = np.argmin(abs(H_Freqs - hi * res))number.append(KEYS[row][col])print(number)return numberfile_name = '附件1.3.wav'
# %%
f = wave.open(file_name, 'rb')
params = f.getparams()
nchannels, sampwidth, framerate, nframes = params[:4]  # 通道数、量化位数（单位字节）、频率、帧数print('nchannels:', nchannels)
print('sampwidth:', sampwidth)
print('framerate:', framerate)
print('nframes:', nframes)strData = f.readframes(nframes)  # 读取音频，字符串格式
if sampwidth == 1:waveData = np.frombuffer(strData, dtype=np.int8)
elif sampwidth == 2:waveData = np.frombuffer(strData, dtype=np.int16)
elif sampwidth == 3:waveData = np.frombuffer(strData, dtype=np.int24)
elif sampwidth == 4:waveData = np.frombuffer(strData, dtype=np.int32)waveData = waveData * 1.0 / (max(abs(waveData)))  # wave幅值归一化
waveData = np.reshape(waveData, [nframes, nchannels])
waveData = waveData[:, 0]
f.close()time = np.arange(0, nframes) * (1.0 / framerate)
plt.figure()
plt.plot(time, waveData)
plt.xlabel("Time(s)")
plt.ylabel("Amplitude")
plt.title("Wavedata")
id('on')  # 标尺，on：有，off:无。
plt.show()
waveData = shape(len(waveData), )
print(waveData)
waveData = filter_bp(waveData, framerate, 680, 1700)
print(waveData)
# waveData[160240:167280] = 0signal_energy(waveData)
result = decode_dtmf(waveData)
print(result)

参考链接 

基于python的DTMF编解码_基于python的dtmf分析-CSDN博客

侵删