Toan-Minh-Duong-Son/vietnamese-music-dataset

Vietnamese Music Dataset

A collection of 4,820 Vietnamese music tracks with matching cover thumbnails and per-track metadata collected from YouTube, packaged as an audiofolder dataset.

Repository structure

Path	Contents	Count
audio/	MP3 audio files, named by YouTube video ID	4,820
images/	PNG cover thumbnails, same IDs as audio/	4,820
data/	Parquet metadata files, one per collection session	31

Metadata schema

Each Parquet file in data/ has the following columns:

Column	Type	Description
id	string	YouTube video ID (matches audio/image filenames)
title	string	Video/song title
artist	string	Channel or performer name
duration	int64	Track duration in seconds
view_count	int64	View count at collection time
like_count	int64	Like count at collection time
upload_date	string	Original upload date (YYYY-MM-DD)
description	string	Video description text
channel_id	string	YouTube channel ID
audio_url	string	URL of the audio file
image_url	string	URL of the thumbnail
subtitle	string	Subtitle/caption text, "None" when absent

Note: audio_url/image_url values reference an n24q02m/vietnamese-music-dataset base path that does not host the files. Resolve files by id inside this repository instead.

Loading

from datasets import load_dataset

ds = load_dataset("Toan-Minh-Duong-Son/vietnamese-music-dataset", split="train")

To load the metadata tables:

from huggingface_hub import snapshot_download
import pandas as pd
import glob

folder = snapshot_download("Toan-Minh-Duong-Son/vietnamese-music-dataset",
                           repo_type="dataset", allow_patterns="data/*")
df = pd.concat(pd.read_parquet(f) for f in sorted(glob.glob(folder + "/data/*.parquet")),
               ignore_index=True).drop_duplicates(subset="id")

Source & provenance

• id values are YouTube video IDs; view_count, like_count, upload_date, and channel_id indicate tracks and thumbnails were collected from YouTube.
• Parquet filenames date the collection between 2025-06-20 and 2025-07-05, across 9 main sessions plus additional "extra" sessions.
• The 31 Parquet files contain ~6,000 rows with duplicates across overlapping sessions; the repository holds 4,820 unique tracks. Deduplicate by id.
• Uploaded by the Toan-Minh-Duong-Son organization. No formal provenance document is included in the repository.

License & usage restrictions

No license has been specified (license: unknown). The recordings, artwork, and metadata originate from third-party YouTube content and are likely protected by copyright. This dataset is made available for research purposes only; users are responsible for complying with applicable law and YouTube's Terms of Service. Do not use commercially.

Intended use & limitations

• Intended for music/audio ML research: classification, retrieval, recommendation.
• Audio quality and loudness vary between sources; loudness normalization is recommended.
• Metadata is incomplete for some tracks (e.g. subtitle = "None"); engagement counts are point-in-time snapshots.
• Session files overlap — always deduplicate by id.

Citation

@misc{vietnamese_music_dataset_2025,
  title        = {Vietnamese Music Dataset},
  author       = {{Toan-Minh-Duong-Son}},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/Toan-Minh-Duong-Son/vietnamese-music-dataset}}
}