PartAI/PSRB

1# Dataset Card for PSRB (Persian Speech Recognition Benchmark) - 1-Hour Sample2 3## Dataset Summary4 5The Persian Speech Recognition Benchmark (PSRB) is a comprehensive dataset designed to evaluate Persian Automatic Speech Recognition (ASR) systems under diverse real-world conditions.  6This 1-hour sample provides a representative subset of the full PSRB corpus, capturing various accents, speech styles, speaker demographics, and acoustic environments.7 8 9## Supported Tasks and Leaderboards10 11- Automatic Speech Recognition (ASR)12 13## Languages14 15- Persian (Farsi)16 17## Dataset Structure18 19### Data Instances20 21Each data instance in this sample is structured as:22 23```json24{25  "audio_path": "file1.wav",26  "text": "می‌گم نمی‌خواین طبق نقشه جلو بریم؟ نقشه رو بی‌خیال غمت نباشه ما مستر رد پا رو داریم. حس بویایی و شم‌ام اشتباه نمی‌کنه. همین الان اشتباه کرده.",27  "audio_duration": 11.88,28  "number_of_speakers": 3,29  "gender": "male",30  "age": "mix",31  "accents": "standard",32  "formality": "informal",33  "semantic_content": "artistic&literary",34  "data_source": "animation",35  "acoustic_environment": "noisy",36  "spontaneous": 137}38```39 40### Data Fields41 42- **`audio_path`**: Path to the `.wav` audio file.43- **`text`**: Transcription of the audio in Persian.44- **`audio_duration`**: Duration of the audio file in seconds.45- **`number_of_speakers`**: Number of speakers present in the clip.46- **`gender`**: Gender of the speaker(s).47- **`age`**: Age category (e.g., child, teen, adult, senior, or mix).48- **`accents`**: Regional accent of the speaker or "standard."49- **`formality`**: Formality level ("formal" or "informal").50- **`semantic_content`**: Semantic topic or domain of the speech (e.g., artistic&literary, technological, medical).51- **`data_source`**: Source type of the data (e.g., animation, podcast, lecture).52- **`acoustic_environment`**: Recording environment (e.g., clean, noisy).53- **`spontaneous`**: 1 if speech is spontaneous, 0 if scripted.54---55 56## Dataset Creation57 58### Curation Rationale59 60The PSRB dataset was created to address the lack of comprehensive Persian ASR resources, covering linguistic diversity (accents, formality) and acoustic variability (clean, noisy, phone calls).61 62### Source Data63 64Data sources include:65- News broadcasts66- Movies and TV shows67- Podcasts68- Lectures69- Audiobooks70- Talk shows71 72Collected from platforms such as Telewebion, Aparat, YouTube, and Iranseda.73 74### Annotations75 76- Transcriptions manually created by expert native Persian speakers.77- Strict two-pass quality control review to ensure consistency and correctness.78- Rich metadata labeling for speaker demographics and speech conditions.79 80### Personal and Sensitive Information81 82- All data was anonymized to protect the identity of participants.83- No personally identifiable information (PII) is present in this dataset.84---85 86## Considerations for Using the Data87 88### Limitations89 90- This sample may not capture the full variability of the complete PSRB corpus.91---92 93## Additional Information94 95 96### Licensing Information97 98This dataset is made available for **research and educational purposes only**.  99 100---101 102## Citation Information103 104If you use this dataset in your research, please cite:105 106```bibtex107@misc{psrb2025,108  title={PSRB: A Comprehensive Benchmark for Evaluating Persian Automatic Speech Recognition Systems},109  author={Nima Sedghiye and Sara Sadeghi and Reza Khodadadi and Farzin Kashani and Omid Aghdaei and Somayeh Rahimi and Mohammad Sadegh Safari},110  year={2025},111  publisher={Part AI Research Center},112  note={Preprint}113}