Large-scale multilingual text-to-speech dataset covering 13 Indian languages with 453,823 utterances. Audio is embedded in Parquet files and playable directly in the dataset viewer.

Languages

Language	Code	Config	Train	Validation
Assamese	`as`	`assamese`	52,251	500
Bengali	`bn`	`bengali`	44,637	500
English	`en`	`english`	12,209	500
Gujarati	`gu`	`gujarati`	16,157	500
Hindi	`hi`	`hindi`	24,211	500
Kannada	`kn`	`kannada`	32,263	500
Malayalam	`ml`	`malayalam`	44,381	499
Marathi	`mr`	`marathi`	28,161	499
Nepali	`ne`	`nepali`	41,320	500
Odia	`or`	`odia`	26,468	500
Punjabi	`pa`	`punjabi`	28,598	500
Tamil	`ta`	`tamil`	53,075	499
Telugu	`te`	`telugu`	43,595	500
Total			447,326	6,497

Usage

from datasets import load_dataset

# Load a specific language
ds = load_dataset("PalakEngineerMaster/Validated_data_TTS", "hindi")

# Access audio
sample = ds["train"][0]
print(sample["audio"])   # {"array": ndarray, "sampling_rate": 16000}
print(sample["text"])    # transcription

Columns

audio: Audio waveform (WAV, 16kHz mono) — playable in viewer
text: Transcription
lang: Language code
emotion: Emotion label
duration: Duration in seconds
utt_id / speaker_id: Utterance and speaker identifiers
source: Data source (rasa, indicvoices_r, indictts)
snr_db: Signal-to-noise ratio (dB)
silence_ratio / clipping_ratio: Audio quality metrics
domain: Domain category

Sources

Downloads last month: 30