Datasets:

mohammadhossein
/

PMG

datasets:
  - mohammadhossein/PMG
language:
  - fa
task_categories:
  - text-to-audio
  - audio-to-audio
  - audio-classification
pretty_name: Persian Music Generation
license: cc-by-nc-4.0
size_categories:
  - 100K<n<1M
dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 32000
    - name: happiness
      dtype: float64
    - name: key
      dtype: string
    - name: spotify_genre
      dtype: string
    - name: popularity
      dtype: float64
    - name: tag
      dtype: string
    - name: artist
      dtype: string
    - name: song
      dtype: string
    - name: spotify
      dtype: string
    - name: start_time
      dtype: float64
    - name: end_time
      dtype: float64
    - name: duration
      dtype: float64
    - name: happiness.1
      dtype: string
    - name: instruments
      dtype: string
    - name: tempo
      dtype: string
    - name: energy
      dtype: string
    - name: instrument_audio
      dtype:
        audio:
          sampling_rate: 32000
    - name: vocal_audio
      dtype:
        audio:
          sampling_rate: 32000
    - name: Caption
      dtype: string
  splits:
    - name: supervised
      num_bytes: 181725679290.516
      num_examples: 67796
    - name: unsupervised
      num_bytes: 539746092542.475
      num_examples: 102215
  download_size: 705294897427
  dataset_size: 721471771832.991
configs:
  - config_name: default
    data_files:
      - split: supervised
        path: data/supervised-*
      - split: unsupervised
        path: data/unsupervised-*

mohammadhossein/PMG

A Persian music dataset with audio clips and rich metadata (artist, genre/tags, instruments, happiness/energy/popularity, tempo, etc.).

Splits Overview

Split	Rows	Total Duration	Mean Dur (s)	Median Dur (s)
`supervised`	67,796	373.83 h (373:49:46)	19.85 s	19.80 s
`unsupervised`	102,215	566.17 h (566:10:18)	19.94 s	20.00 s

Split Descriptions

supervised: Supervised split with rich metadata suitable for text-to-audio or conditioned training.
unsupervised: Unsupervised split focusing on raw audio; metadata may be sparse or partially available.

Data Fields

Below are the columns detected across splits, with guessed descriptions. Types may say varies if they differ by split.

Column	Type	Description
`Caption`	string	Natural-language caption/description of the audio.
`artist`	string	Performer or creator name.
`audio`	dict	Original full mix (accompaniment + vocals)
`duration`	float64	Audio duration (seconds).
`start_time`	float64	Start time (seconds) within a larger source.
`end_time`	float64	End time (seconds) within a larger source.
`energy`	string	Energy score for the clip.
`happiness`	varies (float64, int64)	Happiness/valence score.
`happiness.1`	string	Alternative happiness/valence score (duplicate column in source).
`instrument_audio`	dict	Instrument-only stem (no vocals)
`instruments`	string	List of instruments present (string or list).
`key`	string	Musical key (e.g., C# minor).
`popularity`	varies (float64, int64)	Spotify popularity metric (0–100).
`song`	string	Song/track title.
`spotify`	string	Spotify track/ID/URL or boolean flag.
`spotify_genre`	string	Spotify-derived genre for the track.
`tag`	string	Tag/label assigned to the track/clip.
`tempo`	string	Beats per minute (BPM/tempo).
`vocal_audio`	dict	Vocal-only stem (no accompaniment)

Audio

Primary column: audio_path (🤗 Datasets [Audio]),
- instrument_audio (🤗 Datasets [Audio]),
- vocal_audio (🤗 Datasets [Audio])
Mono @ 32000 Hz
Durations per split are summarized above (Total/Mean/Median/Min/Max).

Licensing

Citation

If you use this dataset, please cite appropriately:

{}