python 语音加窗分帧

from __future__ import division

from scikits.talkbox import segment_axis

import numpy as np

import soundfile as sf

 

def readwav(fn):

    signal, sampleRate = sf.read(fn)

    signal -= np.mean(signal)

    signal /= np.max(np.abs(signal))# Normalize the amplitude

    nframes = len(signal)

    return signal, sampleRate, nframes

 

# smoother version of hanning window

def sqrt_hann(M):

    return np.sqrt(np.hanning(M))

 

# signal: 1D array, returns a 2D complex array

def pro_signal(signal, window = 'hanning', frame_len = 1024, overlap = 512):

    if window == 'hanning':

        # w = np.hanning(frame_len)

        w = sqrt_hann(frame_len)

    else:

        w = window

    y = segment_axis(signal, frame_len , overlap = overlap, end = 'cut')# use cut instead of pad

    y = w * y

    return y

 

# signal is a 2d matrix, n_frames * frame_len

# returns a 2d matrix, n_frames * frame_len /2 + 1

def stft(signal):

    out = np.array([np.fft.rfft(signal[i]) for i in xrange(signal.shape[0])])

    return out

 

# Load the signal

fn = 'lullaby.wav'

signal, sampleRate, nframes = readwav(fn)

print sampleRate, nframes

snd = pro_signal(signal, frame_len = 1024, overlap = 512)

out = stft(snd)

print out.shape