Harbidel/tigrinya-asr-merged

tigrinya-asr-merged A merged Tigrinya speech-recognition dataset, combining and deduplicating: badrex/tigrinya-speech (train pool) google/WaxalNLP config tir_asr (train pool) UBC-NLP/SimbaBench_dataset config asr_test_tir (held-out benchmark test set) Processing Standardized to audio (16kHz mono) and text columns, with a source column tracking origin Unicode NFC-normalized transcripts, empty transcripts dropped Exact-duplicate transcripts removed from the train… See the full description on the dataset page: https://huggingface.co/datasets/Harbidel/tigrinya-asr-merged.