Skip to content
TheChatArchiveThe Chat Archive
gpt-4-0314· OpenAIAi2 WildChat-4.8M (AI2)

Оптимизируй кодimport os import torchaudio import matplotlib.pyplot as plt import torch…

  1. User

    Оптимизируй кодimport os import torchaudio import matplotlib.pyplot as plt import torch import librosa as lr import numpy as np

    Define the vad function

    def vad(audio_data, sample_rate, frame_duration=0.025, energy_threshold=0.6): frame_size = int(sample_rate * frame_duration) num_frames = len(audio_data) // frame_size silence = np.ones(len(audio_data), dtype=bool)

    for idx in range(num_frames):
        start = idx * frame_size
        end = (idx + 1) * frame_size
        frame_energy = np.sum(np.square(audio_data[start:end]))
        if frame_energy > energy_threshold:
            silence[start:end] = False
    
    segments = []
    start = None
    for idx, value in enumerate(silence):
        if not value and start is None:  # Start of speech segment
            start = idx
        elif value and start is not None:  # End of speech segment
            segments.append([start, idx - 1])
            start = None
    
    return segments, silence
    

    directory = '/content/drive/MyDrive/School21/day09/datasets/audio_yes_no/waves_yesno 2'

    Loop through all files in the directory

    for filename in os.listdir(directory): if filename.endswith('.wav'): # Load the audio file and find the speech boundaries audio_data, sample_rate = torchaudio.load(os.path.join(directory, filename)) audio_data_resampled = lr.resample(audio_data.numpy()[0], orig_sr=sample_rate, target_sr=8000) segments, silence = vad(audio_data_resampled, 8000)

        # Resample the silence mask
        silence_resampled = np.interp(np.linspace(0, len(silence), len(audio_data_resampled)), np.arange(len(silence)), silence.astype(float))
    
        # Plot the audio signal and speech boundaries
        fig, ax = plt.subplots(figsize=(14, 4))
        times = np.linspace(0, len(audio_data_resampled) / 8000, len(audio_data_resampled))*2000  # Calculate time values for x-axis
        ax.plot(times, audio_data_resampled, label='Oridgin signal')
    
        # Create a threshold line based on the silence mask
        threshold_line = np.where(silence_resampled, 0, np.max(np.abs(audio_data_resampled)))
        ax.plot(times, threshold_line, color='red', linestyle='-', label='Voice_detected')
    
        plt.title(filename)
        plt.legend()
        plt.show()
    
  2. Assistant
    import os
    import torchaudio
    import matplotlib.pyplot as plt
    import librosa as lr
    import numpy as np
    
    
    def vad(audio_data, sample_rate, frame_duration=0.025, energy_threshold=0.6) -> (list, np.array):
        frame_size = int(sample_rate * frame_duration)
        num_frames = len(audio_data) // frame_size
        silence = np.ones(len(audio_data), dtype=bool)
    
        for idx in range(num_frames):
            start = idx * frame_size
            end = (idx + 1) * frame_size
            frame_energy = np.sum(np.square(audio_data[start:end]))
            if frame_energy > energy_threshold:
                silence[start:end] = False
    
        segments = []
        start = None
        for idx, value in enumerate(silence):
            if not value and start is None:  # Start of speech segment
                start = idx
            elif value and start is not None:  # End of speech segment
                segments.append([start, idx - 1])
                start = None
    
        return segments, silence
    
    
    def plot_audio(audio_data: np.array, sample_rate: int, silence: np.array, title: str):
        silence_resampled = np.interp(np.linspace(0, len(silence), len(audio_data)), np.arange(len(silence)),
                                      silence.astype(float))
    
        fig, ax = plt.subplots(figsize=(14, 4))
        times = np.linspace(0, len(audio_data) / sample_rate, len(audio_data)) * 2000
        ax.plot(times, audio_data, label='Original signal')
    
        threshold_line = np.where(silence_resampled, 0, np.max(np.abs(audio_data)))
        ax.plot(times, threshold_line, color='red', linestyle='-', label='Voice_detected')
    
        plt.title(title)
        plt.legend()
        plt.show()
    
    
    def process_audio_files(directory: str, resample_rate=8000):
        for filename in os.listdir(directory):
            if filename.endswith('.wav'):
                audio_data, sample_rate = torchaudio.load(os.path.join(directory, filename))
                audio_data_resampled = lr.resample(audio_data.numpy()[0], orig_sr=sample_rate, target_sr=resample_rate)
                segments, silence = vad(audio_data_resampled, resample_rate)
    
                plot_audio(audio_data_resampled, resample_rate, silence, filename)
    
    
    if __name__ == "__main__":
        directory = '/content/drive/MyDrive/School21/day09/datasets/audio_yes_no/waves_yesno 2'
        process_audio_files(directory)
    

    I have streamlined the code by:

    1. Breaking down the main loop into separate functions so that it's more modular and easier to understand.
    2. Adding type hints to function definitions to make it clearer what inputs and outputs are being dealt with.

    These changes should make the code simpler to maintain and understand.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.