Datasetpublic

labari-voice/fr-sn-speech-pilot

fr_sn, Senegalese French read-speech pilot Read speech in Senegalese French, in the FLEURS format. The sentences were written in Senegal, about local realities, and read by Senegalese speakers in their own French. For evaluation, not for training. See DATASHEET.md for provenance and intended use. Segments 210 Sentences 105, all covered Speakers 5 (3 female, 2 male) Duration 19.3 min Words 2832 Audio WAV PCM 16-bit, 16 kHz, mono Split single test split… See the full description on the dataset page: https://huggingface.co/datasets/labari-voice/fr-sn-speech-pilot.

sourceHugging Facecc-by-4.0updated 10d agoView on Hugging Face

0likes196downloads

Dataset Card

fr_sn, Senegalese French read-speech pilot

Read speech in Senegalese French, in the FLEURS format. The sentences were written in Senegal, about local realities, and read by Senegalese speakers in their own French. For evaluation, not for training.

See DATASHEET.md for provenance and intended use.


Segments	210
Sentences	105, all covered
Speakers	5 (3 female, 2 male)
Duration	19.3 min
Words	2832
Audio	WAV PCM 16-bit, 16 kHz, mono
Split	single `test` split

Files

audio/test/         segments + metadata.csv
test.tsv            FLEURS metadata
speakers.tsv        speaker table
manifest.jsonl      per-segment measurements
checksums.sha256    audio integrity
DATASHEET.md        datasheet

test.tsv follows the FLEURS schema: no header, seven tab-separated columns.

#	Column
1	`id`, utterance identifier, 1 to 105
2	`file_name`
3	`raw_transcription`, script text with punctuation and casing
4	`transcription`, normalised: lowercased, no punctuation
5	`num_samples`
6	`speaker_id`, see `speakers.tsv`
7	`gender`

An id appears once per speaker who read it. File names encode utterance and speaker as NNNSS.wav. Transcriptions come from the reading script, not from automatic speech recognition.

Speakers

`speaker_id`	Code	Gender	Age	Median F0	Segments
1	`fr-SN_spk01`	FEMALE	twenties	199 Hz	40
2	`fr-SN_spk02`	FEMALE	twenties	205 Hz	31
3	`fr-SN_spk03`	FEMALE	twenties	227 Hz	10
4	`fr-SN_spk04`	MALE	thirties	117 Hz	30
5	`fr-SN_spk05`	MALE	fifties	112 Hz	99

Age follows the Common Voice vocabulary. Identifiers are pseudonymised.

Measurements


Segment duration	3.0 s to 14.6 s, median 5.0 s
Signal-to-noise ratio	median 24.1 dB, minimum 18.8 dB
Peak level	median -10.2 dBFS
Clipped segments	0

Per-segment values are in manifest.jsonl and allow filtering before use.

Limitations

—Volume. Twenty minutes, against roughly twelve hours per language in FLEURS. Not sized for training.
—Statistical power. With 2832 words, an error rate carries a margin of about ± 1.5 point at a 20 % error rate.
—Speaker imbalance, from 10 to 99 segments per voice. Weight accordingly.
—Two single-speaker domains, Telecommunications and Administration.
—Read speech, not spontaneous speech.
—Partition on id rather than on segments, so that every voice reading a given text stays on the same side.

Licence

CC BY 4.0: use, redistribution and derivative works, including commercial use, with attribution.

Recordings were made with narrators under contract in Labari Voice studios, with informed consent for public release. Speaker identifiers are pseudonymised and carry no directly identifying data.

For larger deliveries or other language varieties: sales@labari.dev