import os
import uuid
import soundfile as sf
from datasets import load_dataset

# Settings
dataset_name = "MushanW/GLOBE_V2"  # e.g., "librispeech_asr"
split = "train"  # or "test", "validation", etc.
output_dir = "./wav_output"

# Load dataset
ds = load_dataset(dataset_name, split=split, verification_mode="no_checks")

# Make sure output directory exists
os.makedirs(output_dir, exist_ok=True)

# Iterate through dataset and save audio
for i, sample in enumerate(ds):
    audio = sample["audio"]
    samples = audio["array"]
    sample_rate = audio["sampling_rate"]
    
    filename = f"{uuid.uuid4()}.wav"
    filepath = os.path.join(output_dir, filename)
    
    # Save as .wav file
    sf.write(filepath, samples, sample_rate)

    if (i + 1) % 100 == 0:
        print(f"Saved {i+1} files...")

print(f"Done! Saved {len(ds)} files to '{output_dir}'")