ParsVoice: A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis

📋 Overview

ParsVoice is the largest high-quality Persian speech dataset designed specifically for text-to-speech (TTS) applications. The dataset addresses the critical gap in Persian speech technologies by providing a comprehensive corpus with speaker diversity and audio quality comparable to major English corpora.

🎯 Key Features

• 1,804 hours of high-quality speech data
• 470+ unique speakers with diverse characteristics
• Multi-speaker TTS optimized content
• High-quality audio-text alignment using automated pipeline
• Naturalness MOS: 3.6/5
• Speaker Similarity SMOS: 4.0/5

📊 Dataset Statistics

Metric	Value
Total Audio Duration	1,804 hours (high-quality subset)
Raw Audio Processed	3,526 hours
Number of Speakers	470+
Source Material	2,000 audiobooks
Language	Persian (Farsi)
Audio Format	WAV
Sample Rate	22.05 kHz

🔬 Research Paper

This dataset is introduced in our paper:

ParsVoice: A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis

Mohammad Javad Ranjbar Kalahroodi, Heshaam Faili, Azadeh Shakery

University of Tehran

📄 Read the full paper: arXiv:2510.10774

Abstract

Existing Persian speech datasets are typically smaller than their English counterparts, which creates a key limitation for developing Persian speech technologies. We address this gap by introducing ParsVoice, the largest Persian speech corpus designed specifically for text-to-speech (TTS) applications. We created an automated pipeline that transforms raw audiobook content into TTS-ready data, incorporating components such as a BERT-based sentence completion detector, a binary search boundary optimization method for precise audio-text alignment, and audio-text quality assessment frameworks tailored to Persian. The pipeline processes 2,000 audiobooks, yielding 3,526 hours of clean speech, which was further filtered into a 1,804-hour high-quality subset suitable for TTS, featuring more than 470 speakers. To validate the dataset, we fine-tuned XTTS for Persian, achieving a naturalness Mean Opinion Score (MOS) of 3.6/5 and a Speaker Similarity Mean Opinion Score (SMOS) of 4.0/5, demonstrating ParsVoice's effectiveness for training multi-speaker TTS systems.

📦 Dataset Access

⚠️ Important Notice:

A representative subset of the ParsVoice dataset is currently available for preview and research purposes.

Full dataset access will be granted after the paper is accepted for publication.

For early access requests or collaboration inquiries, please contact the authors.

🏗️ Dataset Structure

Each sample in the dataset contains:

{
    "audio": Audio feature,
    "text": str,              # Transcript of the audio
    "speaker_id": str,        # Unique speaker identifier
    "duration": float,        # Audio duration in seconds
    "speaker_gender": str,    # Speaker gender (M/F)
}

💻 Usage

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("MohammadJRanjbar/ParsVoice")

# Access a sample
sample = dataset['train'][0]
print(f"Text: {sample['text']}")
print(f"Duration: {sample['duration']} seconds")
print(f"Speaker: {sample['speaker_id']}")

Example: Training a TTS Model

from datasets import load_dataset
import torch

# Load dataset
dataset = load_dataset("MohammadJRanjbar/ParsVoice", split="train")

# Your TTS training code here
for sample in dataset:
    audio = sample["audio"]["array"]
    text = sample["text"]
    # Process for TTS training

🔧 Data Processing Pipeline

The ParsVoice dataset was created using an automated pipeline that includes:

1. BERT-based sentence completion detector for text segmentation
2. Binary search boundary optimization for precise audio-text alignment
3. Quality assessment frameworks tailored for Persian speech
4. Multi-stage filtering to ensure high-quality TTS data

🎯 Applications

This dataset is suitable for:

• Text-to-Speech (TTS) synthesis
• Voice cloning and conversion
• Speaker recognition
• Speech enhancement
• Persian language model development
• Multi-speaker synthesis research

📈 Benchmark Results

We validated the dataset by fine-tuning XTTS for Persian:

Metric	Score
Naturalness (MOS)	3.6/5
Speaker Similarity (SMOS)	4.0/5

These results demonstrate ParsVoice's effectiveness for training high-quality multi-speaker TTS systems.

📜 Citation

If you use this dataset in your research, please cite:

@article{ranjbar2024parsvoice,
  title={ParsVoice: A Large-Scale Multi-Speaker Persian Speech Corpus for Text-to-Speech Synthesis},
  author={Ranjbar Kalahroodi, Mohammad Javad and Faili, Heshaam and Shakery, Azadeh},
  journal={arXiv preprint arXiv:2510.10774},
  year={2024}
}

📄 License

This dataset is released under [specify license - e.g., CC BY-NC 4.0, MIT, etc.]. Please refer to the LICENSE file for more details.

👥 Authors

• Mohammad Javad Ranjbar Kalahroodi - University of Tehran
• Heshaam Faili - University of Tehran
• Azadeh Shakery - University of Tehran

📧 Contact

For questions, issues, or collaboration opportunities:

• Open an issue on this repository
• Email: [contact email]
• Project Page: [if available]

🙏 Acknowledgments

We thank all contributors and the University of Tehran for supporting this research.

📊 Dataset Card

• Curated by: Mohammad Javad Ranjbar Kalahroodi, Heshaam Faili, Azadeh Shakery
• Language: Persian (Farsi)
• License: [To be specified]
• Paper: arXiv:2510.10774

---

Note: This is a research dataset. Please ensure compliance with applicable laws and ethical guidelines when using this data.