Adaptability of ASR Models on Low-Resource Language: A Comparative Study of Whisper and Wav2Vec-BERT on Bangla
It addresses the problem of developing efficient speech recognition systems for low-resource languages like Bangla, but is incremental as it applies existing methods to new data.
This study compared two state-of-the-art ASR models, Whisper and Wav2Vec-BERT, on Bangla, a low-resource language, finding that Wav2Vec-BERT outperformed Whisper in all key metrics like Word Error Rate and required fewer computational resources.
In recent years, neural models trained on large multilingual text and speech datasets have shown great potential for supporting low-resource languages. This study investigates the performances of two state-of-the-art Automatic Speech Recognition (ASR) models, OpenAI's Whisper (Small & Large-V2) and Facebook's Wav2Vec-BERT on Bangla, a low-resource language. We have conducted experiments using two publicly available datasets: Mozilla Common Voice-17 and OpenSLR to evaluate model performances. Through systematic fine-tuning and hyperparameter optimization, including learning rate, epochs, and model checkpoint selection, we have compared the models based on Word Error Rate (WER), Character Error Rate (CER), Training Time, and Computational Efficiency. The Wav2Vec-BERT model outperformed Whisper across all key evaluation metrics, demonstrated superior performance while requiring fewer computational resources, and offered valuable insights to develop robust speech recognition systems in low-resource linguistic settings.