adjaysagar/high-quality-tts

High-Quality TTS (Sarvam)

Multilingual Indian language TTS dataset (Sarvam/Bulbul) in Hugging Face format. One config per language for subset download.

Downloading language-wise subsets

Each language is a separate configuration (subset), so you can load only one language (like ai4bharat/Rasa):

from datasets import load_dataset, get_dataset_config_names

# Single language (only that language's parquet is downloaded)
ds = load_dataset("adjaysagar/high-quality-tts", "bengali", trust_remote_code=True)  # Bengali
ds["train"]  # Dataset for that language

# List available language configs
get_dataset_config_names("adjaysagar/high-quality-tts")  # e.g. ['bengali', 'telugu', ...]

# All languages (DatasetDict: config -> Dataset)
ds_all = load_dataset("adjaysagar/high-quality-tts", trust_remote_code=True)

Dataset structure

Column	Description
audio	22.05 kHz audio (WAV)
text	Transcript (from Samanantar)
language	Language name

Language and metadata statistics

Duration (h) = total audio duration per language. Config name is used in load_dataset(..., "bengali").

Language	Config	Utterances	Duration (h)	GB	Unique speakers	States	Districts
Bengali	bengali	300,000	427.37	63.20	0	0	0
English	english	69,209	87.90	13.00	0	0	0
Gujarati	gujarati	188,021	245.66	36.33	0	0	0
Kannada	kannada	300,000	385.40	57.00	0	0	0
Malayalam	malayalam	300,000	380.07	56.21	0	0	0
Marathi	marathi	300,000	415.71	61.48	0	0	0
Odia	odia	69,883	113.04	16.72	0	0	0
Punjabi	punjabi	70,000	122.31	18.09	0	0	0
Tamil	tamil	300,000	431.44	63.80	0	0	0
Telugu	telugu	300,000	385.72	57.04	0	0	0
Total	—	2,197,113	2994.62	442.87	—	—	—