convert_dataset.py

import torchaudio
import torch
import os
from datasets import load_dataset
import numpy as np

# Function to convert a single audio sample to WAV file
def convert_audio_sample_to_wav(audio_data, sample_rate, output_file):
    """
    Converts a single audio sample (tensor) to a WAV file.

    Parameters:
    - audio_data (torch.Tensor): The audio data as a tensor.
    - sample_rate (int): The sample rate of the audio data.
    - output_file (str): The path to save the output WAV file.
    """
    try:
        # Scale audio data to the range of int16
        audio_tensor = audio_data / torch.max(torch.abs(audio_data))
        audio_tensor = audio_tensor * 32767
        audio_tensor = audio_tensor.short()

        # Ensure the audio_tensor is 2D (channels x samples)
        if audio_tensor.ndim == 1:
            audio_tensor = audio_tensor.unsqueeze(0)  # Add channel dimension

        # Save the audio tensor to a WAV file
        torchaudio.save(output_file, audio_tensor, sample_rate)

        print(f"Saved {output_file}")

    except Exception as e:
        print(f"Error saving {output_file}: {e}")

# Main script
if __name__ == "__main__":
    # Specify the dataset path and split
    dataset_path = "/home/vikrant/Conversational-AI-Model/embedding_vocoder/English_Accent_DataSet_Local/westbrook___english_accent_data_set/default/0.0.0"
    split = "train"  # Or "validation" or "test"

    # Load the dataset split using Hugging Face Datasets
    try:
        dataset = load_dataset("arrow", data_dir=os.path.join(dataset_path, split), split=split)
    except Exception as e:
        print(f"Error loading dataset: {e}")
        exit()

    # Define the output directory for saving WAV files
    output_wav_dir = "output_wavs"

    # Create the output directory if it doesn't exist
    if not os.path.exists(output_wav_dir):
        os.makedirs(output_wav_dir)

    # Iterate over each example in the dataset
    for index, example in enumerate(dataset):
        try:
            # Extract relevant information from the example
            audio_id = example['audio_id']
            audio_data = torch.tensor(example['audio']['array']).float()  # Get audio data as a tensor
            sample_rate = example['audio']['sampling_rate']  # Get sample rate

            # Construct the output file path
            output_file = os.path.join(output_wav_dir, f"{audio_id}.wav")

            # Convert and save the audio sample to a WAV file
            convert_audio_sample_to_wav(audio_data, sample_rate, output_file)

        except Exception as e:
            print(f"Error processing example {index}: {e}")

    print("Conversion complete.")