PalakEngineerMaster/Processed_TTS_Multilingual_Data

1---2language:3  - as4  - bn5  - en6  - gu7  - hi8  - kn9  - ml10  - mr11  - ne12  - or13  - pa14  - ta15  - te16license: cc-by-4.017task_categories:18  - text-to-speech19  - automatic-speech-recognition20size_categories:21  - 100K<n<1M22tags:23  - indic24  - multilingual25  - tts26  - speech27---28 29# Processed TTS Multilingual Data30 31Validated and quality-checked multilingual speech datasets for TTS training, covering 12+ Indian languages.32 33## Datasets Included34 35| Subset | Samples | Hours | Description |36|---|---|---|---|37| `indic_voices_r` | 239,684 | 548.8h | Indic Voices_R — IVR recordings |38| `rasa` | 201,509 | 361.2h | RASA — read speech (wiki, conv, book, news) |39| `indictts_iitm` | 155,236 | 253.6h | Indic TTS (IIT Madras) — studio TTS recordings at 48kHz |40| **Total** | **596,429** | **1,163.6h** | |41 42## Languages43 44Assamese (as), Bengali (bn), English (en), Gujarati (gu), Hindi (hi), Kannada (kn), Malayalam (ml), Marathi (mr), Nepali (ne), Odia (or), Punjabi (pa), Tamil (ta), Telugu (te)45 46## Structure47 48```49├── indic_voices_r/50│   ├── metadata.csv51│   └── audio/{lang}/*.wav52├── rasa/53│   ├── metadata.csv54│   └── audio/{lang}/*.wav55└── indictts_iitm/56    ├── metadata.csv57    └── audio/{lang}/*.wav58```59 60## Schema (metadata.csv)61 62Each subset has a `metadata.csv` with these columns:63 64| Field | Description |65|---|---|66| `file_name` | Relative path to audio file (e.g., `audio/bn/file.wav`) |67| `text` | Transcript text |68| `lang` | ISO 639-1 language code |69| `speaker_id` | Speaker identifier |70| `duration` | Audio duration in seconds |71| `source` | Original data source |72| `emotion` | Emotion label |73| `domain` | Text domain (wiki, conv, book, news, etc.) |74| `snr_db` | Signal-to-noise ratio in dB |75| `silence_ratio` | Fraction of silent frames |76| `clipping_ratio` | Fraction of clipped samples |77 78## Quality Checks Applied79 80All data has been validated through a 4-check pipeline:811. **SNR + Silence + Duration** — reject low SNR (<10dB), excess silence (>35%), out-of-range duration (<1.5s or >30s), clipping (>1%)822. **Speaking Rate** — reject abnormal speaking rates (<2 or >25 chars/sec)833. **Text Normalization** — Unicode NFC normalization applied844. **Audio Corruption** — reject empty, all-zeros, NaN/Inf, DC offset >0.185 86## Usage87 88```python89from datasets import load_dataset90 91# Load a specific subset92ds = load_dataset(93    "PalakEngineerMaster/Processed_TTS_Multilingual_Data",94    data_dir="rasa",95    split="train",96)97 98# Access a sample99sample = ds[0]100print(sample["text"])101# audio is at sample["file_name"]102```103 104## Audio Format105 106- Format: WAV107- Sample rate: 16kHz (Indic Voices_R, RASA) / 48kHz (Indic TTS IIT M)108- Channels: mono109