detect_stutter.py

import librosa
import numpy as np
import tensorflow as tf
from pydub import AudioSegment
from keras.models import load_model
import os

model_rep = load_model('/home/mansi/anaconda3/beproject/stutter_det/models/best_model_rep.h5')
model_pro = load_model('/home/mansi/anaconda3/beproject/stutter_det/models/best_model_pro.h5')

def detect_prolongation(mfcc): 
    s = 0
    for m in mfcc:
        y = model_pro.predict(m.reshape(1,2,44,1), batch_size=1)
        y = np.around(y,decimals=2)
        if y[0][0] > 0.5:
            s += y[0][0]
    p_sev = s/len(mfcc)*100
    return p_sev

def detect_repetition(mfcc):
    s = 0
    for m in mfcc:
        y = model_rep.predict(m.reshape(1,13,44,1), batch_size=1)
        y = np.around(y,decimals=2)
        if y[0][0] > 0.5:
            s += y[0][0]
    r_sev = s/len(mfcc)*100
    return r_sev

def detect_stutter(audio):
    sound_file = AudioSegment.from_wav(audio)
    audio_chunks = sound_file[::1000]
    ps = 0
    rs = 0
    mfcc_arr_p = []
    mfcc_arr_r = []
    for i, chunk in enumerate(audio_chunks):
        chunkfile = "chunks_test/chunk{0}.wav".format(i)
        chunk.export(chunkfile, format="wav")
        y, sr = librosa.load(chunkfile)
        mfcc = np.array(librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13))
        
        if mfcc.shape[0] == 13 and mfcc.shape[1] == 44:
            a = []
            a.append(mfcc)
            mfcc_arr_r.append(a)
            b = []
            b.append(mfcc[0])
            b.append(mfcc[12])
            mfcc_arr_p.append(b)
            
    mfcc_arr_r = np.array(mfcc_arr_r)  
    mfcc_arr_p = np.array(mfcc_arr_p)    
    
    mfcc_arr_r.reshape(mfcc_arr_r.shape[0], 13, 44, 1)
    mfcc_arr_p.reshape(mfcc_arr_p.shape[0], 2, 44, 1)
    
    p_sev = detect_prolongation(mfcc_arr_p)
    r_sev = detect_repetition(mfcc_arr_r)
    
    o_sev = (p_sev+r_sev)/2
    
    return p_sev, r_sev, o_sev

if __name__== "__main__":
    common = '/home/mansi/anaconda3/beproject/stutter_det/demo_audios'
    arr1 = os.listdir(common)
    for a in arr1:
        print('\n'+a)
        arr2 = os.listdir(common+'/'+a)
        for b in arr2:
            if b.endswith('.wav'):
                print('\n'+b)
                p_sev, r_sev, o_sev = detect_stutter(common+'/'+a+'/'+b)
                print('Prolongation % : '+str(p_sev))
                print('Repetition % : '+str(r_sev))
                print('Overall stutter % : '+str(o_sev))