PersianVox_NM

PersianVox_NM is a high-quality Persian speech dataset derived from article readings by a single female speaker. This subset is sourced from Nasle Mana Magazine, a publication dedicated to producing accessible content for the visually impaired.

📚 Dataset Summary

Language: Persian (Farsi)
Speaker: Single female voice
Total Duration: 94.55 hours
Recording Source: Articles from Nasle Mana
Domain: Literary and informational prose
Alignment Checked With:
- In-house Whisper model fine-tuned from whisper-large-v3
- In-house NVIDIA FastConformer fine-tuned from NVIDIA FastConformer-Hybrid Large (fa)
- Public wav2vec2_v3

🔑 Key Descriptions

The dataset entries are stored as dictionaries with the following keys:

🧩 Common Keys

Key	Description
`audio_filepath`	Path to the audio file (usually relative to the dataset root)
`duration`	Length of the audio in seconds
`text`	Normalized transcript with normalization and space correction, used to calculate metrics
`text_no_preprocessing`	Transcript before any text cleaning or normalization
`text_normalized`	Cleaned and normalized transcript, suitable for ASR training
`score`	Average confidence score indicating alignment quality (log space); low scores (< -2) are dropped
`start_abs`	Average absolute amplitude at the beginning of the audio segment
`end_abs`	Average absolute amplitude at the end of the audio segment
`tokens`	Number of BPE tokens in the transcript

🤖 Model-Specific Keys

For each ASR model, the following metrics are included with a model prefix (e.g., nvidia_fastconformer, wav2vec2_v3, faster_whisper_large):

Key Template	Description
`pred_text_{model}`	Raw predicted transcript from the model
`pred_text_normalized_{model}`	Normalized prediction for metric computation
`cer_{model}`	Character Error Rate
`wer_{model}`	Word Error Rate
`ins_rate_{model}`	Insertion rate (extra characters)
`del_rate_{model}`	Deletion rate (missed characters)
`sub_rate_{model}`	Substitution rate (wrong characters)

🔄 Licensing

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) License.

You must credit the creator when using or sharing this dataset.

🧠 Intended Use and Limitations

This dataset is intended for:

Training and evaluating ASR systems in Persian
Research in model robustness via ASR scoring
Automatic speech-text alignment studies

🚫 Ethical Notice

Please do not use this dataset to clone or imitate the speaker’s voice for malicious or unethical purposes.

✉️ Citation

To cite this dataset:

@misc{persianvox_nm,
  title = {PersianVox_NM: A 90+ hour single-speaker Persian speech dataset},
  author = {Saeedreza Zouashkiani},
  year = 2025,
  url = {https://naslemana.com/},
  note = {Derived from Nasle Mana Magazine, licensed under CC BY 4.0}
}

Datasets:

saeedzou
/

PersianVox_NM

You need to agree to share your contact information to access this dataset