BCN2BRNO Automatic speech recognition system for Albayzin 2022 Speech to Text Challenge

The software is based on the development of Automatic Speech Recognition systems for the Albayzin 2022 Challenge. We trained and evaluated both hybrid systems and those based on end-to-end models. We also investigated the use of self-supervised learning speech representations from pre-trained models and their impact on ASR performance (as opposed to training models directly from scratch). Additionally, we also applied the Whisper model in a zero-shot fashion, postprocessing its output to fit the required transcription format. On top of tuning the model architectures and overall training schemes, we improved the robustness of our models by augmenting the training data with noises extracted from the target domain. Moreover, we applied rescoring with an external LM on top of N-best hypotheses to adjust each sentence score and pick the single best hypothesis. All these efforts lead to a significant WER reduction. Our single best system and the fusion of selected systems achieved 16.3% and 13.7% WER respectively on RTVE2020 test partition, i.e. the official evaluation partition from the previous Albayzin challenge