plp语音特征提取python代码

以下是使用Python实现PLP语音特征提取的示例代码：

import numpy as np
import scipy.signal as signal

def pre_emp(signal, coeff=0.95):
    # 预加重
    return signal[1:] - coeff * signal[:-1]

def framing(signal, fs, frame_len=0.025, frame_shift=0.01):
    # 分帧
    frame_size = int(frame_len * fs)
    frame_step = int(frame_shift * fs)
    signal_len = len(signal)
    num_frames = int(np.ceil(float(np.abs(signal_len - frame_size)) / np.abs(frame_step)))
    pad_signal_len = num_frames * frame_step + frame_size
    pad_signal = np.zeros((pad_signal_len,))
    pad_signal[:signal_len] = signal
    indices = np.tile(np.arange(0, frame_size), (num_frames, 1)) + np.tile(np.arange(0, num_frames * frame_step, frame_step), (frame_size, 1)).T
    frames = pad_signal[indices.astype(np.int32, copy=False)]
    return frames

def hamming_window(frame):
    # 加窗
    return frame * np.hamming(len(frame))

def power_spectrum(frame, nfft):
    # 计算功率谱
    return np.square(np.abs(np.fft.rfft(frame, nfft)))

def mel_filter_bank(nfilt, nfft, fs):
    # 计算Mel滤波器组
    low_freq_mel = 0
    high_freq_mel = (2595 * np.log10(1 + (fs / 2) / 700))  # 把 Hz 转换为 Mel
    mel_points = np.linspace(low_freq_mel, high_freq_mel, nfilt + 2)
    hz_points = (700 * (10**(mel_points / 2595) - 1))  # 把 Mel 转换为 Hz
    bin = np.floor((nfft + 1) * hz_points / fs)
    fbank = np.zeros((nfilt, int(np.floor(nfft / 2 + 1))))
    for m in range(1, nfilt + 1):
        f_m_minus = int(bin[m - 1])
        f_m = int(bin[m])
        f_m_plus = int(bin[m + 1])
        for k in range(f_m_minus, f_m):
            fbank[m - 1, k] = (k - bin[m - 1]) / (bin[m] - bin[m - 1])
        for k in range(f_m, f_m_plus):
            fbank[m - 1, k] = (bin[m + 1] - k) / (bin[m + 1] - bin[m])
    return fbank

def mfcc(signal, fs, nfilt=40, nfft=512):
    # 提取MFCC特征
    pre_emphasized_signal = pre_emp(signal)
    frames = framing(pre_emphasized_signal, fs)
    frames *= np.hamming(frames.shape[1])
    NFFT = np.power(2, np.ceil(np.log2(frames.shape[1])))
    mag_frames = np.absolute(np.fft.rfft(frames, NFFT))
    pow_frames = ((1.0 / NFFT) * (np.square(mag_frames)))
    fb = mel_filter_bank(nfilt, NFFT, fs)
    feat = np.dot(pow_frames, fb.T)
    feat = np.where(feat == 0, np.finfo(float).eps, feat)
    feat = 20 * np.log10(feat)
    feat = signal.lfilter([1], [1, -0.97], feat) # 可选步骤，对MFCC系数进行预加重
    feat = dct(feat, type=2, axis=1, norm='ortho')[:, :13] # 只保留前13个系数
    return feat

上述代码中，`pre_emp`函数实现预加重，`framing`函数实现分帧，`hamming_window`函数实现加窗，`power_spectrum`函数实现计算功率谱，`mel_filter_bank`函数实现计算Mel滤波器组，`mfcc`函数实现提取MFCC特征。其中，`nfilt`表示滤波器组数量，`nfft`表示FFT点数，`type`表示DCT类型（1或2），`axis`表示进行DCT的轴，`norm`表示是否进行正交归一化。