karya/sanmati-stories

Sanmati Stories

Sanmati Stories contains spoken readings of story sentences in six Indian languages. A worker reads one sentence aloud, and the recording is that reading. This is not the leading-questions dataset.

text is the sentence that was read, not the whole story. story_id is the story that contains that sentence.

What is in each row

Column	Description
text	Sentence that was read.
tar_file	Repository path of the uncompressed tar shard that holds the WAV.
audio_in_tar	Member name of the WAV inside that tar.
sha256	SHA-256 hex digest of the WAV bytes.
num_bytes	WAV size in bytes.
lang	Language code: bh, bn, hi, ml, mr, or te.
story_id	Story that contains the sentence.
sentence_id	Sentence identifier from the export.
worker_id	Worker identifier from the export.
duration	Source duration, in milliseconds.
speech_present	Source silence flag.

Audio files are stored in uncompressed tar shards next to the parquet metadata:

• data/{lang}/train-00000-of-NNNNN.parquet
• data/{lang}/train-00000-of-NNNNN.tar

Each parquet row points at one WAV through tar_file and audio_in_tar.

Hugging Face's audio player may not play members packed inside tar files. Open the sibling tar and read audio_in_tar.

Languages

Choose a configuration to load one language. Each configuration has a single train split. There is no held-out evaluation split.

Config	Language	Examples
bh	Bhojpuri	383,880
bn	Bengali	700,895
hi	Hindi	293,337
ml	Malayalam	1,095,037
mr	Marathi	559,156
te	Telugu	714,985

There are 3,747,290 usable clips in total. Usable clips are WAV files that join to a sentence.

Loading the data

from datasets import load_dataset

bengali = load_dataset("karya/sanmati-stories", "bn", split="train")
print(bengali[0]["text"])
print(bengali[0]["story_id"], bengali[0]["sentence_id"])
print(bengali[0]["tar_file"], bengali[0]["audio_in_tar"])

Access

Access is manually approved. Accept the conditions on the dataset page before downloading files.

Acknowledgments

We gratefully acknowledge the communities and individuals who contributed these recordings and transcripts.

License

This work is licensed under a KPL BY-NC-SA-FS 1.0 license. This license allows reusers to distribute, remix, adapt, build upon, and incorporate into software systems, the material in any medium or format for noncommercial purposes only, and only so long as attribution is given to the creator. If you remix, adapt, build upon, or incorporate into software systems, you must license the modified material, including material generated by the software system, under identical terms, and license the software system under the GNU General Public License.

See LICENSE in this repository and the full text at karya-inc/kpl.

Commercial Use

If you are interested in using this dataset for commercial purposes, please contact us. Profits from commercial licensing will be distributed as royalties to the community members who contributed to this dataset.

Contact

For questions, issues, or contributions, open an issue on the dataset repository or contact us directly.