A Case Study on Filtering for End-to-End Speech Translation
This work addresses data quality issues for researchers and practitioners in speech translation, but it is incremental as it applies a basic filtering method to an existing problem.
The study tackled the problem of low-quality, noisy parallel corpora in speech translation by applying a simple filtering technique to create a cleaner dataset, resulting in an average improvement of 4.65 BLEU score for multilingual-to-English speech translation models.
It is relatively easy to mine a large parallel corpus for any machine learning task, such as speech-to-text or speech-to-speech translation. Although these mined corpora are large in volume, their quality is questionable. This work shows that the simplest filtering technique can trim down these big, noisy datasets to a more manageable, clean dataset. We also show that using this clean dataset can improve the model's performance, as in the case of the multilingual-to-English Speech Translation (ST) model, where, on average, we obtain a 4.65 BLEU score improvement.