sayiwen/aopcode-sarvinaz-v1-clean

AOPCODE Sarvinaz V1 Clean - Uyghur Text-to-Speech Dataset

Dataset Description

This is a high-quality, professionally curated Uyghur text-to-speech dataset provided by AOPCODE. The dataset contains 22,052 clean audio samples with corresponding Uyghur text transcriptions, specifically processed for TTS model training.

Dataset Highlights

• 🎯 Pure Uyghur Content: All entries contain only Uyghur text (no English letters or digits)
• 🔊 High Quality Audio: Professional voice recordings at 22kHz sampling rate
• 📝 Clean Transcriptions: Carefully processed and normalized text
• 🛠️ Ready for Training: Pre-split into train/validation sets (90%/10%)
• 📊 Standardized Format: Compatible with popular TTS frameworks

Dataset Statistics

Split	Samples
Train	~19,847
Validation	~2,205
Total	22,052

• Audio Format: WAV, 22kHz, mono
• Text Language: Uyghur (维吾尔语/ئۇيغۇر تىلى)
• Character Set: 35 unique characters (Uyghur script + punctuation)
• Domain: Conversational speech, social interactions, shopping, greetings

Data Fields

Each sample contains:

• audio: Audio file (WAV format, 22kHz)
• sentence: Normalized Uyghur text transcription

Character Set

The dataset uses a clean, standardized Uyghur character set:

[' ', 'ئ', 'ا', 'ب', 'ت', 'ج', 'خ', 'د', 'ر', 'ز', 'س', 'ش', 'غ', 'ف', 'ق', 'ك', 'ل', 'م', 'ن', 'و', 'ى', 'ي', 'پ', 'چ', 'ژ', 'ڭ', 'گ', 'ھ', 'ۆ', 'ۇ', 'ۈ', 'ۋ', 'ې', 'ە', '…']

Data Processing

The dataset has been carefully processed to ensure maximum quality:

1. Text Normalization: Unicode normalization (NFKC)
2. Character Filtering: Removed entries with English letters and digits
3. Punctuation Cleaning: Standardized punctuation handling
4. Quality Control: Verified audio-text alignment
5. Deduplication: Removed duplicate entries

Usage Example

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("sayiwen/aopcode-sarvinaz-v1-clean")

# Access training data
train_data = dataset["train"]
validation_data = dataset["validation"]

# Example sample
sample = train_data[0]
audio = sample["audio"]
text = sample["sentence"]

Applications

This dataset is ideal for:

• Text-to-Speech (TTS) model training
• Speech Synthesis research
• Uyghur Language Processing applications
• Voice Assistant development
• Educational Tools for Uyghur language learning

License & Commercial Use

This dataset is provided under a Commercial License.

• ✅ Commercial use permitted with proper licensing
• 📄 License required for commercial applications
• 🤝 Custom licensing available for enterprise clients

Contact & Licensing

For commercial licensing, custom datasets, or technical support:

• 🌐 Website: aopcode.com
• 📧 Contact: Available through our website
• 💼 Enterprise Solutions: Custom data collection and processing services available

Citation

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

@dataset{aopcode_sarvinaz_v1_clean,
  title={AOPCODE Sarvinaz V1 Clean - Uyghur TTS Dataset},
  author={AOPCODE},
  year={2024},
  url={https://huggingface.co/datasets/sayiwen/aopcode-sarvinaz-v1-clean},
  publisher={AOPCODE}
}

Quality Assurance

• ✅ Audio quality verified (22kHz, clear recordings)
• ✅ Text accuracy manually reviewed
• ✅ Character set standardized
• ✅ Train/validation split optimized
• ✅ Ready for immediate use in TTS training

Version History

• v1.0: Initial clean release with 22,052 samples
• Processed from original AOPCODE Sarvinaz dataset
• Filtered and normalized for optimal TTS training

---

Note: This is a commercial dataset. For licensing inquiries and technical support, please visit aopcode.com.