Hugging Face's logo

Datasets:
JosueG
/
adja-speech-asr-tts

Tasks:
Automatic Speech Recognition
Text-to-Speech
Modalities:
Text
Formats:
parquet
Languages:
Adioukrou
French
Size:
1K - 10K
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
text
stringlengths
5
119
audio
dict
Lé mí wɛ yi nyamɔn nyɛ ŋ mɔnŋujemɛ
{"array":[0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.(...TRUNCATED)
Kpɔnnɔ ŋ du kpoŋu aŋu
{"array":[0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.(...TRUNCATED)
ŋ kpɔ mɔ́ wɔ ale hunnɔ
{"array":[0.0,-0.000030517578125,-0.0001220703125,-0.00030517578125,-0.00054931640625,-0.00091552734(...TRUNCATED)
Tom can Marie can, wowo dele bɔbɔ lɔ mɛ ò
{"array":[0.0,-0.0001220703125,-0.000335693359375,-0.00079345703125,-0.001434326171875,-0.0023193359(...TRUNCATED)
Zili le dɔmɛzi nuxu xo ɔ̀
{"array":[0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.(...TRUNCATED)
Le enu jɔjɔ lɔo yi ɔ do yi dɔ yi wo xo nuxu ahan
{"array":[0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.(...TRUNCATED)
ŋ bumɔ́ wɔ a kpɔɛ alo wɔ a nya
{"array":[0.0,0.0,-0.00006103515625,-0.0001220703125,-0.000244140625,-0.000396728515625,-0.000640869(...TRUNCATED)
ŋ ɖu nyidin ŋuɖuɖu fɛn du
{"array":[0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.(...TRUNCATED)
Eshi vɔ le anyi nuŋlwɛ ci mɛ
{"array":[0.0,0.0,0.0,0.0,-0.000030517578125,-0.00006103515625,-0.000091552734375,-0.0001220703125,-(...TRUNCATED)
Na yi ji anyi hún kpɔ
{"array":[0.0,-0.000030517578125,-0.0001220703125,-0.00030517578125,-0.000579833984375,-0.0009765625(...TRUNCATED)
End of preview. Expand in Data Studio

Adja Speech Dataset for ASR and TTS

This is the canonical public Adja speech dataset for the May 2026 thesis release. It is intended for automatic speech recognition, text-to-speech, and speech pipeline experiments.

Components

adja_speech_orpheus_48khz

This component duplicates the Orpheus speech source from JosueG/adja-tts-orpheus. The source repository is treated as read-only provenance; this dataset repo is the public canonical release surface.

The component is useful for ASR because each row pairs Adja text with speech, and useful for TTS because the same pairs can train or evaluate speech synthesis models.

Release split policy:

  • No new train/dev/test split is created for the public dataset.
  • The full Orpheus corpus is published as the single Hugging Face train partition because that is the source dataset shape.

HF partition:

  • train: 1,597 rows

Schema:

  • text: normalized transcription text
  • audio: speech audio stored from the Orpheus source at 48 kHz

Provenance And Quality Note

The processed derivative JosueG/adja-tts-mms-ready is not used as the canonical public source after release review found bad/noisy playback on Hugging Face. This repo goes back to the Orpheus source data instead.

No existing experiment dependency repo is renamed, moved, or edited by this release.

Related Thesis Artifacts

  • Code and result archive: FrejusGdm/cs-thesis-may-2026
  • Existing canonical MT dataset: JosueG/french-adja-parallel-corpus

Citation

If you use this dataset, cite:

Josue Godeme. 2026. CS Thesis May 2026: French-Adja MT and Adja Speech Experiments.
Downloads last month
72
Number of rows:
1,597
Total file size:
1.34 GB

Models trained or fine-tuned on JosueG/adja-speech-asr-tts

Collection including JosueG/adja-speech-asr-tts