Text-Utilization for Encoder-dominated Speech Recognition Models
This work provides practical guidance for researchers and practitioners building faster speech recognition systems by simplifying text-data integration.
The paper explores efficient methods for using text-only data to improve encoder-dominated speech recognition models, finding that simple configurations like random duration models outperform complex alternatives and that larger encoders with smaller decoders can match or exceed larger-decoder architectures on LibriSpeech.
This paper investigates efficient methods for utilizing text-only data to improve speech recognition, focusing on encoder-dominated models that facilitate faster recognition. We provide a comprehensive comparison of techniques to integrate text-only data, including modality matching and dynamic downsampling to reach text-level representations within the encoder. Our experiments on the LibriSpeech corpus show that a larger encoder with a smaller decoder can equal or surpass the performance of architectures with larger decoders. We demonstrate that simple configurations, such as random duration models, are often more effective than complex alternatives, significantly simplifying the training pipeline. All code and recipes are made publicly available.