TrixyZ/client/hotword_detection.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import sounddevice as sd
from scipy.io.wavfile import write
import librosa
import numpy as np
from tensorflow.keras.models import load_model
import pyaudio

class HotwordDetector:
    def __init__(self, model_path):
        self.model_path = model_path
        self.LoadModel()
        self.rec_duration = 3
        
    def LoadModel(self):
        #self.interpreter = tf.lite.Interpreter(model_path=self.model_path)
        #self.interpreter.allocate_tensors()
        #self.input_details = self.interpreter.get_input_details()
        #self.output_details = self.interpreter.get_output_details()
        self.audio = pyaudio.PyAudio()
        self.model = load_model(self.model_path)
        self.mic_stream = self.audio.open(
            format=pyaudio.paInt16,
            channels=1,
            rate=44100,
            input=True,
            output=True,
            frames_per_buffer=4096)
        

    #def record_audio(self):
    #    p = pyaudio.PyAudio()
    #    stream = p.open(format=pyaudio.paInt16,
    #                    channels=1,
    #                    rate=16000,
    #                    input=True,
    #                    frames_per_buffer=self.chunk_size)
    #    frames = []
    #    for _ in range(int(16000 / self.chunk_size * 1)):
    #        data = stream.read(self.chunk_size)
    #        frames.append(np.frombuffer(data, dtype=np.int16))
    #    stream.stop_stream()
    #    stream.close()
    #    p.terminate()
    #    return np.concatenate(frames)

    #def predict(self, audio_data):
    #    input_data = np.expand_dims(audio_data, axis=0).astype(np.float32)
    #    self.interpreter.set_tensor(self.input_details[0]['index'], input_data)
    #    self.interpreter.invoke()
    #    output_data = self.interpreter.get_tensor(self.output_details[0]['index'])
    #    return output_data[0][0]

    def detect_hotword(self):
        if True:
            myrecording = sd.rec(int(3 * 44100), samplerate=44100, channels=2)
            sd.wait()
            write("tmp.wav", 44100, myrecording)
            audio, sample_rate = librosa.load("tmp.wav")
            mfcc = librosa.feature.mfcc(y=audio, sr=sample_rate, n_mfcc=40)
            mfcc_processed = np.mean(mfcc.T, axis=0)
            prediction = self.model.predict(np.expand_dims(mfcc_processed, axis=0))
        else:
            data = self.mic_stream.read(4096)
            data_float = np.fromstring(data , dtype=np.float16)
            data_np = np.array(data_float , dtype='d')
            mfcc = librosa.feature.mfcc(data_np.flatten(), 44100, n_mfcc=40)
            mfcc_processed = np.mean(mfcc.T, axis=0)
            prediction = self.model.predict(np.expand_dims(mfcc_processed, axis=0))
        print(prediction[0])
        print(prediction[0][1])
        print(prediction[0][2])
        if prediction[0][1] > 0.90:
            print(f"Wake Word 'Trixy' Detected")
            print("Confidence:", prediction[0][1])
        elif prediction[0][2] > 0.90:
            print(f"Wake Word 'SystemCommand' Detected")
            print("Confidence:", prediction[0][2])
        else:
            print("Wake Word NOT Detected")
            print("Confidence:", prediction[0][0])