AfriQA
CollaborativeCreation of an open-access evaluation dataset for answering questions in African languages.
- Question-réponse
- Langues africaines
- Corpus
Active
Research and data
GalsenAILab
What the community builds, and what comes out of it. Projects on one side, the corpora they produce on the other.
Projects
What the community builds
Work supported by GalsenAI and collaborative initiatives it participates in.
-
-
Masakhane
CollaborativeResearch community working on NLP for African languages.
- NLP
- Traduction automatique
- Recherche ouverte
Active -
Waxal
GalsenAIA voice recognition project for Wolof.
- Reconnaissance vocale
- Langues africaines
- Corpus
Active -
Adama
GalsenAIA Wolof text-to-speech model.
- Synthèse vocale
- Langues africaines
Active -
Aya
CollaborativeAn open-science initiative by Cohere For AI to build a state-of-the-art multilingual language model.
- Modèles de langue
- Multilingue
- Science ouverte
Active -
Traduction automatique
GalsenAIA machine translation project between French and Wolof.
- Traduction automatique
- Langues africaines
Active
Datasets
What comes out of it
Senegalese languages remain poorly endowed with exploitable data. This section gathers corpora produced by the community and those it recommends, detailing their domain, languages, volume, and license.
Filter the catalog
13 datasets listed
-
Synthèse vocale
Cleaned and denoised version of the female voice from the Wolof TTS corpus. Annotations were revised to remove special characters, emojis and foreign characters, and clips judged to be of insufficient quality were discarded. Represents 18 h 41 min of usable speech.
- Wolof
- Volume
- 19,947 audio clips · 18 h 41 min
- License
- CC BY 4.0
Open on Hugging Face
-
Reconnaissance vocale
Automatic speech recognition dataset for Wolof, built by bringing together four existing corpora and their transcriptions. Designed for training ASR models, it is the largest transcribed audio collection in Wolof published by the community.
- Wolof
- Volume
- 35,075 audio clips · 6.5 GB
- License
- Apache 2.0
Open on Hugging Face
-
Synthèse vocale
Speech synthesis corpus recorded by two native Wolof speakers, one male and one female voice, each having read more than 20,000 sentences. Collected by Baamtu Datamation as part of the AI4D African Language Program, it is the reference basis for speech synthesis in Wolof.
- Wolof
- Volume
- 40,010 audio clips · 4.6 GB
- License
- CC BY 4.0
Open on Hugging Face
-
Traduction automatique
Aggregation of the various Wolof-French translation sources available within the community, brought together in a single format with traceability of each pair's origin. This is the largest Wolof-French parallel corpus published by GalsenAI.
- Wolof
- Français
- Volume
- 98,345 sentence pairs
- License
- License not specified by the producer , check the terms of reuse before use
Open on Hugging Face
-
Traduction automatique
Small English-Wolof parallel corpus, designed for rapid evaluation and for bootstrapping translation models into English.
- Wolof
- Anglais
- Volume
- 7,405 sentence pairs
- License
- License not specified by the producer , check the terms of reuse before use
Open on Hugging Face
-
Traduction automatique
Reference evaluation set for machine translation, covering 200 languages including Wolof. Because the same sentences are translated into every language, it makes it possible to compare a Wolof model against results obtained on better-resourced languages.
- Wolof
- Français
- Anglais
- Multilingue
- Volume
- 200 languages
- License
- CC BY-SA 4.0
Open on Hugging Face
-
Traduction automatique
French-Wolof parallel corpus aligned sentence by sentence, each pair retaining a reference to its original source. Used to train the community translation models.
- Wolof
- Français
- Volume
- 17,777 sentence pairs
- License
- License not specified by the producer , check the terms of reuse before use
Open on Hugging Face
-
Reconnaissance vocale
Transcribed speech corpus devoted to agriculture in the three most widely spoken languages of Senegal. The recordings bring together farmers, agricultural advisors and agrifood business managers, through interactive radio programmes, focus groups, voice messages and interviews. Project funded by the Lacuna Fund.
- Wolof
- Pulaar
- Sérère
- Volume
- 125 h of transcribed speech, 35 h of them verified
- License
- CC BY 4.0
Open on Zenodo
-
Traduction automatique
Parallel translation corpus in the news domain, covering 21 African languages including Wolof, with French and English as pivot languages. Also produced by the Masakhane collective.
- Wolof
- Français
- Anglais
- Multilingue
- Volume
- 21 African languages
- License
- CC BY-NC 4.0
Open on Hugging Face
-
Reconnaissance d'entités
Named entity recognition corpus covering 20 African languages, including Wolof. Produced by the Masakhane collective, to which GalsenAI members contribute. Annotations cover people, places, organisations and dates.
- Wolof
- Multilingue
- Volume
- 20 African languages
- License
- AFL 3.0
Open on Hugging Face
-
Classification audio
Voice command dataset covering Wolof, Pulaar and Serer, annotated by keyword. Each audio clip is paired with one of the words in the target vocabulary, making it a resource suited to keyword spotting and to voice interfaces in national languages.
- Wolof
- Pulaar
- Sérère
- Volume
- Between 10,000 and 100,000 clips
- License
- CreativeML OpenRAIL-M
Open on Hugging Face
-
Reconnaissance vocale
Correction of the Wolof portion of the WAXAL corpus published by Google, whose audio files and transcriptions suffered from a misalignment that made the resource unusable. The community regenerated the transcriptions and realigned the whole set, making the corpus usable again.
- Wolof
- Volume
- 1,042 audio clips · 6.1 GB
- License
- License not specified by the producer , check the terms of reuse before use
Open on Hugging Face
-
Corpus textuel
Monolingual text corpus in Wolof, intended for the pre-training and adaptation of language models. Serves as the basis for the Wolof language modelling work carried out within the community.
- Wolof
- Volume
- 52,706 sentences
- License
- License not specified by the producer , check the terms of reuse before use
Open on Hugging Face
A project to share?
If you are working on Senegalese languages, on an open corpus, or on a model that the community could reuse, write to us.