README.md

metadata

language:
  - fa
multilinguality:
  - monolingual
size_categories:
  - 10K<n<100K
task_categories:
  - automatic-speech-recognition
task_ids:
  - speech-recognition
pretty_name: ParsVoice - Persian Speech Recognition Dataset
tags:
  - persian
  - farsi
  - speech
  - audio
  - asr
license: mit
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: transcription
      dtype: string
    - name: transcription_with_punctuation
      dtype: string
    - name: speaker_id
      dtype: string
    - name: book_id
      dtype: string
    - name: is_complete
      dtype: bool
    - name: duration_seconds
      dtype: float64
    - name: snr_db
      dtype: float64
    - name: quality_score
      dtype: float64
    - name: segment_type
      dtype: string
    - name: was_smart_trimmed
      dtype: bool
    - name: speaker_embedding
      list: float64
    - name: embedding_confidence
      dtype: float64
  splits:
    - name: train
      num_bytes: 323754892
      num_examples: 916
  download_size: 318080676
  dataset_size: 323754892

ParsVoice - Persian Speech Recognition Dataset

Dataset Description

ParsVoice is a high-quality Persian (Farsi) speech recognition dataset created from audiobooks. The dataset features clean, professionally narrated speech with accurate transcriptions.

Dataset Statistics

• Total Samples: 2,624
• Total Books: 10
• Total Narrators: 5
• Language: Persian (Farsi)
• Sampling Rate: 16 kHz
• Audio Format: WAV

Features

• audio: High-quality 16kHz audio segments
• transcription: Raw transcription text
• transcription_with_punctuation: Transcription with restored punctuation
• speaker_id: Encoded speaker/narrator identifier
• book_id: Encoded book identifier
• is_complete: Whether the segment contains a complete sentence
• duration_seconds: Length of audio segment
• snr_db: Signal-to-noise ratio
• quality_score: Overall quality score (0-100)
• segment_type: Type of segment processing applied
• was_smart_trimmed: Whether smart trimming was applied

Quality Assurance

• All segments pass quality filters for SNR, background music detection, and distortion
• Smart trimming applied to remove excess silence while preserving speech content
• Only high-quality segments with clear transcriptions included

Privacy

• Book titles and narrator names are encoded for privacy
• Original content identification is not possible from the dataset

Usage

from datasets import load_dataset

dataset = load_dataset("MohammadJRanjbar/ParsVoice")

# Access audio and transcription
for sample in dataset['train']:
    audio = sample['audio']
    text = sample['transcription']
    # Use for training ASR models

License

MIT License - See LICENSE file for details.

Citation

If you use this dataset, please cite:

@dataset{parsvoice2025,
  title={ParsVoice: A High-Quality Persian Speech Recognition Dataset},
  author={Mohammad Ranjbar},
  year={2025},
  url={https://huggingface.co/datasets/MohammadJRanjbar/ParsVoice}
}