BabelSpeech/40hours_Indonesian_Colloquial_ASR_Speech_Dataset

BabelSpeech 50-Hour Indonesian Colloquial ASR Speech Dataset

• Contains 50 hours of Indonesian colloquial ASR speech data, aligned with natural, everyday Indonesian communication patterns.
• Metadata is stored in a separate JSON file, including audio path, duration, transcript confidence, signal-to-noise ratio (SNR), and DNSMOS. More metadata fields may be added in future updates.
• Covered domains: technology, entertainment, travel, education, daily life, and others.
• Data quality: Each batch is spot-checked by native speakers, with overall accuracy > 98%, making it suitable for ASR model fine-tuning.
• Sampling rate: All audio is sampled at 16 kHz.

Data Filtering and Quality Control Workflow

1. High-quality public data sources are collected via collaboration with international channels, covering multiple real-world domains.
2. All data undergoes secondary manual screening to remove noisy or contextually chaotic segments, ensuring clear audio quality.
3. The data is then passed through an automated processing pipeline for denoising, enhancement, and structured cleaning.
4. Automated audio-quality assessment and multi-dimensional scoring mechanisms are applied to select the best-quality audio segments.
5. Multi-model cross-validation is used to improve transcription quality and reduce systematic bias.
6. Finally, native speakers perform spot checks to ensure that the overall transcription accuracy remains above 98%.

Metadata Format

{
    "filename": "youtube_id_spk009_d5m7BbsbKow_202509_0001_00.wav",
    "relative_path": "wav/youtube_id_spk009_d5m7BbsbKow_202509_0001_00.wav",
    "duration": 4.7,
    "confidence": 1.0,
    "text": " aku lagi ada di istanbul di bandara terbarunya aku baru pertama kali kesini juga",
    "snr": 55.97792053222656,
    "dnsmos": 4.101025
}

Data Distribution

• SNR Distribution

  

• DNSMOS Distribution

  

• Duration Distribution

  

• Confidence Distribution

  

Contact

If you have any questions or would like to obtain more high-quality Indonesian speech data, please feel free to contact our team via email: babelspeechai@gmail.com