voice.py

import whisper
import sounddevice as sd
import numpy as np
import gc

model = whisper.load_model("base")  # Load the Whisper model

sd.default.dtype = 'float16'  # Set the default data type for recording
sample_rate = sd.default.samplerate = 16000  # Set the default sample rate for recording

def record_audio(duration=5):
    """
    Record audio from the microphone for a given duration.
    
    :param duration: Duration in seconds to record audio.
    :return: Recorded audio as a NumPy array.
    """
    audio_data = sd.rec(int(duration * sample_rate), samplerate=sample_rate, channels=1, dtype='float32')
    sd.wait()  # Wait until recording is finished
    return audio_data  

def transcribe_audio(model, audio_data):
    """
    Transcribe audio data using the Whisper model.
    
    :param audio_data: NumPy array containing the recorded audio.
    :return: Transcribed text.
    """
    # Whisper expects the audio to be in a specific format, so we need to convert it to the correct format
    audio_data = np.squeeze(audio_data)  # Remove extra dimensions if necessary
    result = model.transcribe(audio_data, language='en')
    return result['text']

def wait_for_silence(model, interval, threshold=0.7, silence_duration=2):
    """
    Wait for silence in the audio stream and stop recording.
    
    :param model: Whisper model for transcription
    :param interval: Time interval to check for silence (in seconds)
    :param threshold: Silence threshold (amplitude below this is considered silence)
    :param silence_duration: Duration of silence required to stop (in seconds)
    :return: Transcribed text from the recording
    """
    frames = []
    silence_count = 0
    print("Starting recording...")
    
    while True:
        audio_data = record_audio(duration=interval)
        frames.append(audio_data)
        
        # Calculate RMS amplitude of the current audio chunk
        rms = np.sqrt(np.mean(np.square(audio_data)))
        
        if rms < threshold:
            silence_count += interval
            if silence_count >= silence_duration:
                print("Silence detected, stopping recording...")
                break
        else:
            silence_count = 0
    
    return transcribe_audio(model, np.concatenate(frames))