AstralZander/ASP_yoruba

# ASP_yoruba We are solving a speech recognition problem for the Yoruba language. **As promising models for training , the following were considered:** 1. facebook/wav2vec2-xls-r-300m 2. openai/whisper-small 3. Ayoola/cdial-yoruba-test (Ayoola/cdial-yoruba-test) **We use datasets for training**: 1. google/fleurs (FLEURS: Few-shot Learning E…) 2. bibleTTS (Bible TTS) 3. mozilla-foundation/common_voice_12_0 4. Lagos-NWU (repo.sadilar.org) WER (Word error rate) is used as the main evaluation metric. **WER**: 1. *67.8866* **Dataset**: google/fleurs **Model**: openai/whisper-small **Source**: steja/whisper-small-yoruba 2. *57.8640* **Dataset**: google/fleurs + mozilla-foundation/common_voice_12_0 **Model**: facebook/wav2vec2-xls-r-300m **Source**: huggingface.co