Holuwasege/OWE-Voice-Evaluation

Evaluation of Owe-voice dataset on open-source monolingual and multilingual model (NCAIR1/Yoruba-ASR & MMS) # Òwe-Voice: ASR Evaluation on Yoruba Proverb Speech This repository contains code for evaluating automatic speech recognition (ASR) models on Òwe-Voice, a Yoruba proverb speech dataset. ## Models Evaluated - NCAIR1/Yoruba-ASR-v0.1 - facebook/mms-1b-all ## Metrics - Word Error Rate (WER) - Tone Error Rate (TER) ## Evaluation Setup - Zero-shot evaluation - CPU-only inference - Sampling rate: 16kHz ## Repository Structure ... ## Citation If you use this work, please cite: