AppOnly.py

#!/usr/bin/env python
# coding: utf-8

# In[1]:


class_names = ['cothe', 'khong', 'nguoi', 'toi', 'va']

import librosa
import numpy as np
import os
import math
from sklearn.cluster import KMeans
import hmmlearn.hmm

def get_mfcc(file_path):
    y, sr = librosa.load(file_path) # read .wav file
    hop_length = math.floor(sr*0.010) # 10ms hop
    win_length = math.floor(sr*0.025) # 25ms frame
    # mfcc is 12 x T matrix
    mfcc = librosa.feature.mfcc(
        y, sr, n_mfcc=12, n_fft=1024,
        hop_length=hop_length, win_length=win_length)
    # substract mean from mfcc --> normalize mfcc
    mfcc = mfcc - np.mean(mfcc, axis=1).reshape((-1,1)) 
    # delta feature 1st order and 2nd order
    delta1 = librosa.feature.delta(mfcc, order=1)
    delta2 = librosa.feature.delta(mfcc, order=2)
    # X is 36 x T
    X = np.concatenate([mfcc, delta1, delta2], axis=0) # O^r
    # return T x 36 (transpose of X)
    return X.T # hmmlearn use T x N matrix


# In[2]:


#loadmodels
import pickle

model = {}
for key in class_names:
    name = f"models\model_{key}.pkl"
    with open(name, 'rb') as file:
        model[key] = pickle.load(file)


# In[ ]:





# In[ ]:





# In[3]:


from tkinter import messagebox
import winsound

from pydub import AudioSegment

import ffmpeg

#Thay đổi threshold dựa vào tạp âm, càng ồn thì threshold càng lớn
def detect_leading_silence(sound, silence_threshold=-42.0, chunk_size=10):
    '''
    sound is a pydub.AudioSegment
    silence_threshold in dB
    chunk_size in ms

    iterate over chunks until you find the first one with sound
    '''
    trim_ms = 0 # ms

    assert chunk_size > 0 # to avoid infinite loop
    while sound[trim_ms:trim_ms+chunk_size].dBFS < silence_threshold and trim_ms < len(sound):
        trim_ms += chunk_size

    return trim_ms

def record():
    import pyaudio
    import wave
    from base64 import b64decode

    CHUNK = 1024
    FORMAT = pyaudio.paInt16
    CHANNELS = 1
    RATE = 22050
    RECORD_SECONDS = 2
    WAVE_OUTPUT_FILENAME = "record.wav"

    p = pyaudio.PyAudio()

    stream = p.open(format=FORMAT,
                    channels=CHANNELS,
                    rate=RATE,
                    input=True,
                    frames_per_buffer=CHUNK)

    frames = []

    for i in range(0, int(RATE / CHUNK * RECORD_SECONDS)):
        data = stream.read(CHUNK)
        frames.append(data)

    stream.stop_stream()
    stream.close()
    p.terminate()

    wf = wave.open(WAVE_OUTPUT_FILENAME, 'wb')
    wf.setnchannels(CHANNELS)
    wf.setsampwidth(p.get_sample_size(FORMAT))
    wf.setframerate(RATE)
    wf.writeframes(b''.join(frames))
    wf.close()

def play():    
    filename = 'record.wav'
    winsound.PlaySound(filename, winsound.SND_FILENAME)
    
def playtrimmed():    
    filename = 'trimmed.wav'
    winsound.PlaySound(filename, winsound.SND_FILENAME)

def predict():
    #Trim silence
    sound = AudioSegment.from_file("record.wav", format="wav")

    start_trim = detect_leading_silence(sound)
    end_trim = detect_leading_silence(sound.reverse())

    duration = len(sound)    
    
    trimmed_sound = sound[start_trim:duration-end_trim]    
    trimmed_sound.export("trimmed.wav", format="wav")
    
    #Predict
    record_mfcc = get_mfcc("trimmed.wav")
    scores = [model[cname].score(record_mfcc) for cname in class_names]
    pred = np.argmax(scores)
    messagebox.showinfo("result", class_names[pred])


# In[4]:


import tkinter as tk

window = tk.Tk()
window.geometry("300x200")
window.title("Speech recognition")

frame0 = tk.Frame(master=window)
frame0.pack()

frame1 = tk.Frame(master=window)
frame1.pack()

frame2 = tk.Frame(master=window)
frame2.pack()

label = tk.Label(master=frame0, text="Speech recognition")
label.pack(padx=5, pady=10)

btn_record = tk.Button(master=frame1, width=13, height=2, text="record", command=record)
btn_record.pack(side=tk.LEFT, padx=5, pady=5)

btn_playback = tk.Button(master=frame1, width=13, height=2, text="playback", command=play)
btn_playback.pack(side=tk.LEFT, padx=5, pady=5)

btn_predict = tk.Button(master=frame2, width=13, height=2, text="trim & predict", command=predict)
btn_predict.pack(side=tk.LEFT, padx=5, pady=5)

btn_playback = tk.Button(master=frame2, width=13, height=2, text="playbacktrimmed", command=playtrimmed)
btn_playback.pack(side=tk.LEFT, padx=5, pady=5)



window.mainloop()


# In[ ]: