The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
Amharic ASR MLOps workshop data
Prepared 16 kHz mono WAV recordings and the manifest used by the Ethiopia-Training workshop.
This repository is a DVC remote. Clone the workshop repo and run
uv run dvc pull. You don't need to download files from here by hand.
Sources and licences
- FLEURS Amharic (
google/fleurs,am_et), licensed CC BY 4.0. Conneau et al., FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech, 2022. - ALFFA Amharic read speech (OpenSLR 25), licensed MIT. Speech data acquired from Solomon T. and Martha Y.; prepared by Michael Melese (Addis Ababa University), Laurent Besacier and Elodie Gauthier (GETALP LIG), and Million Meshesha (Addis Ababa University), as stated in the corpus README.
Audio was converted to 16 kHz mono 16-bit WAV. Transcripts are unchanged from the sources. Every file's provenance (source, original id, licence, sha256) is in the manifest.
Model transcripts
The repository also holds what four pinned models (Whisper small and large-v3-turbo, Omnilingual ASR CTC-300M and LLM-300M) wrote for the 200-recording evaluation subset, clean and through a simulated phone line, with the hardware and software each was made on. Participants score and compare them without needing a GPU.
- Downloads last month
- 127