sound_plt.py

#!/usr/bin/env python
import sys
import numpy as np
import matplotlib.pyplot as plt
import wave
import scipy.signal as signal

def downsampling(conversion_rate, data, fr) :
    decimation_sampleNum = conversion_rate-1
    nyqF = (fr/conversion_rate)/2.0
    cF = (fr/conversion_rate/2.0-500.)/nyqF
    taps = 511
    b = signal.firwin(taps, cF)
    data = signal.lfilter(b, 1, data)
    down_data = []
    for i in range(0, len(data), decimation_sampleNum+1):
        down_data.append(data[i])

    return (down_data, int(fr/conversion_rate))

def smoothing(input, window):
    output = []
    for i in range(input.shape[0]):
        if i < window:
            output.append(np.mean(input[:i+window+1]))
        elif i > input.shape[0] - 1 - window:
            output.append(np.mean(input[i:]))
        else:
            output.append(np.mean(input[i-window:i+window+1]))
    return np.array(output)

def to_db(data, N):
    pad = np.zeros(N//2)
    pad_data = np.concatenate([pad, data, pad])
    rms = np.array([np.sqrt((1/N) * (np.sum(pad_data[i:i+N]))**2) \
                    for i in range(len(data))])
    with np.errstate(divide='ignore'):
            db = 20 * np.log10(rms)
    return db

def sound_plt(wav_fname):
        wave_file = wave.open(wav_fname,"rb") 
        fr = wave_file.getframerate()
        nframes = wave_file.getnframes()
        data = wave_file.readframes(nframes) 
        data = np.frombuffer(data, dtype= "int16")
        down_rate = 2
        down_fr = int(fr / (down_rate * 1000))
        down_data, down_fr = downsampling(down_fr , data, fr)

        N = int(fr / 42)
        db = to_db(down_data, N)

        time = np.arange(0, db.shape[0] / down_fr, 1 / down_fr) / 60 / down_rate
        sm_db = smoothing(db, 1000)

        sm_db_x = [i for i in sm_db if i >= 20]
        db_mean = np.mean(sm_db_x)
        db_max = (np.max(sm_db_x))
        db_max_time = time[np.argmax(sm_db)]

        sm_db_s = [i for i in sm_db if i < 20]
        silent_time = len(sm_db_s) / down_fr / down_rate

        plt.rcParams['font.family'] = 'sans-serif'
        plt.rcParams['font.sans-serif'] = ['IPAPGothic', 'VL PGothic']

        fig = plt.figure(figsize=(16, 8), dpi=100, facecolor='tan', tight_layout=True)
        ax = fig.add_subplot(111, fc='w', xlabel='分(min)', ylabel='音量(dB)')
        ax.set_title("スーパーカブ １話 音量推移")
        ax.plot(time, sm_db, label='signal')

        num = fr
        b = np.ones(num)/num
        sm_db2 = np.convolve(sm_db, b, mode='same')
        ax.plot(time, sm_db2 , 'r', label='moving average')
        
        ax.set_ylim(20, 95)
        boxdic = {
            "facecolor" : "tan",
            "edgecolor" : "w",
            "boxstyle" : "Round",
            "linewidth" : 1
        }
        ax.axhline(y=db_mean , color='g',linestyle='dashed', linewidth=1)
        ax.text(-2, db_mean, "average:" + str(round(db_mean,1)) + "dB", size=10)
        ax.axvline(x=db_max_time , color='g',linestyle='dashed', linewidth=1)
        ax.text(db_max_time-1, db_max+1, \
                "max:" + str(round(db_max,1)) + \
                'dB(' + str(round(db_max_time,1)) +'min)', \
                size=10)
        ax.text(0.01, 0.99, \
                "silent time: " + str(int(silent_time)) + 'sec' \
                , verticalalignment='top', transform=ax.transAxes, bbox=boxdic)
        
        ax.grid()
        plt.legend()
        fig.savefig('ret.png', facecolor=fig.get_facecolor())

        plt.show()

if __name__ == '__main__':
   _, wav_fname = sys.argv
   sound_plt(wav_fname)