metadata
license: other
language:
- hi
- en
task_categories:
- text-to-speech
- automatic-speech-recognition
size_categories:
- 100K<n<1M
pretty_name: 500hrs Audio Data
500hrs Audio Data
~535 hours of speech audio for Hindi and English. All audio is resampled to 44,100 Hz.
Overview
| Stat | Value |
|---|---|
| Total samples | 260,244 |
| Total duration | 535.0 hours |
| Total audio files | 260,586 |
| Uncompressed size | 158.42 GB |
Splits
Training set (95%)
- Samples: 247,232
- Duration: 508.3 hours (30,499.1 minutes)
- Hindi: 134,483 samples (54.4%)
- English: 112,749 samples (45.6%)
- Hinglish: 0 samples (0.0%)
Validation set (5%)
- Samples: 13,012
- Duration: 26.7 hours (1,599.7 minutes)
- Hindi: 7,078 samples (54.4%)
- English: 5,934 samples (45.6%)
- Hinglish: 0 samples (0.0%)
Dataset sources
Hindi
| Source | Samples processed | Hours |
|---|---|---|
| ai4bharat/indicvoices_r | 26,318 | 71.84 |
| ai4bharat/Rasa | 12,112 | 21.37 |
| SPRINGLab/IndicTTS-Hindi | 11,718 | 24.81 |
| Shekharmeena/Single-speaker-female-hindi-high-quality | 15,435 | 38.34 |
| Ritwika03/hindi_syspin | 43,133 | 101.35 |
| skywalker290/Hindi_TTS_M | 17,797 | 40.49 |
| manish92596/roopa-ca-4000-hinglish-num | 4,022 | 6.57 |
| manish92596/vishal-al-4000-hinglish-num | 4,022 | 6.29 |
| manish92596/vishal-al-7000-hinglish | 7,004 | 9.16 |
| Hindi total | 141,561 | 320.2 |
English
| Source | Samples processed | Hours |
|---|---|---|
| MrDragonFox/Elise | 1,194 | 2.55 |
| ylacombe/expresso | 11,614 | 10.92 |
| ylacombe/podcast_fillers_processed | 17,161 | 49.34 |
| amaai-lab/DisfluencySpeech | 4,500 | 8.56 |
| deepvk/NonverbalTTS | 3,641 | 9.35 |
| IISc/SPICOR_English | 50,127 | 98.74 |
| IISc/IndicTIMIT_Hindi | 30,446 | 35.30 |
| English total | 118,683 | 214.8 |
Top 10 sources by sample count
- IISc/SPICOR_English: 47,591 samples
- Ritwika03/hindi_syspin: 40,979 samples
- IISc/IndicTIMIT_Hindi: 28,897 samples
- ai4bharat/indicvoices_r: 25,026 samples
- skywalker290/Hindi_TTS_M: 16,911 samples
- ylacombe/podcast_fillers_processed: 16,343 samples
- Shekharmeena/Single-speaker-female-hindi-high-quality: 14,656 samples
- ai4bharat/Rasa: 11,507 samples
- SPRINGLab/IndicTTS-Hindi: 11,145 samples
- ylacombe/expresso: 11,045 samples
Contents of this repo
- voice_data.zip — Zipped dataset (audio + manifests). Contains:
audio/— WAV files (44,100 Hz)train.jsonl— Training manifest (path, text, duration, language)val.jsonl— Validation manifest
Preparation details
- Sample rate: 44,100 Hz
- Stratified split: 5% validation by language