load_script.py

from datasets import load_dataset
import numpy as np

# Path to your new parquet directory
parquet_dir = "/opt/gpfs/home/leying/data/data/DeepASMR_dataset/parquet"

print("Loading dataset from sharded Parquet files...")
# We use a wildcard to load all shards at once
dataset = load_dataset("parquet", data_files=f"{parquet_dir}/*.parquet", split="train")

print(f"Dataset loaded successfully!")
print(f"Total number of utterances: {len(dataset)}")

# --- Test a random sample ---
sample_idx = 0 
sample = dataset[sample_idx]

print("\n--- Sample Check ---")
print(f"Transcript: {sample['transcript']}")
print(f"Speaker ID: {sample['speaker_id']}")
print(f"Language:   {sample['language']}")

# This part tests if the audio bytes actually decode into a waveform
audio_array = sample['audio']['array']
sampling_rate = sample['audio']['sampling_rate']

print(f"Audio Shape: {audio_array.shape}")
print(f"Sampling Rate: {sampling_rate} Hz")
print(f"Duration: {len(audio_array) / sampling_rate:.2f} seconds")

# Optional: Check if the data is non-zero (proves it's not just silence/empty)
if np.abs(audio_array).max() > 0:
    print("Status: Audio data is valid and non-silent.")
else:
    print("Status: Warning - Audio data appears to be empty/silent.")oooo rclone size deepasmr:deepasmr-transfer-2026/DeepASMR-DB