Datasetpublic

djelia/bambara-tts-waxal

bambara-tts-waxal Bambara studio speech from the WAXAL corpus — 1,926 recordings, 16 hours, 8 speakers, 44.1 kHz mono. Load from datasets import load_dataset ds = load_dataset("djelia/bambara-tts-waxal", "google_waxal", split="train") Splits: train, validation, test. Fields Field Description audio 44.1 kHz mono text Transcript speaker_id Speaker identifier (8 distinct) gender Speaker gender locale Locale code id Record… See the full description on the dataset page: https://huggingface.co/datasets/djelia/bambara-tts-waxal.

sourceHugging Faceupdated 2mo agoView on Hugging Face

0likes69downloads

Dataset Card

bambara-tts-waxal

Bambara studio speech from the WAXAL corpus — 1,926 recordings, 16 hours, 8 speakers, 44.1 kHz mono.

Load

python

from datasets import load_dataset

ds = load_dataset("djelia/bambara-tts-waxal", "google_waxal", split="train")

Splits: train, validation, test.

Fields

Field	Description
`audio`	44.1 kHz mono
`text`	Transcript
`speaker_id`	Speaker identifier (8 distinct)
`gender`	Speaker gender
`locale`	Locale code
`id`	Record identifier

Notes

Transcripts cover 746 of the 1,926 recordings. The rest carry the string "nan" rather than an empty value, so filter on it explicitly:

python

transcribed = ds.filter(lambda row: row["text"] != "nan")

The untranscribed recordings are still usable for speaker modelling or as unlabelled audio.

Speakers appear across all three splits. If you need a speaker-disjoint evaluation, build your own split on speaker_id.