T5lephone: Bridging Speech and Text Self-supervised Models for Spoken Language Understanding via Phoneme level T5
This work addresses the granularity mismatch in aligning speech and text models for spoken language understanding, offering a computationally efficient solution that improves performance on specific tasks like spoken question answering and speech translation.
The paper tackled the problem of aligning speech and text models in spoken language understanding by exploring different tokenization strategies for pretrained language models, and introduced T5lephone, a variant of T5 pretrained on phonemicized text, which achieved state-of-the-art results on NMSQA and outperformed T5 with other units on spoken question answering and speech translation tasks.
In Spoken language understanding (SLU), a natural solution is concatenating pre-trained speech models (e.g. HuBERT) and pretrained language models (PLM, e.g. T5). Most previous works use pretrained language models with subword-based tokenization. However, the granularity of input units affects the alignment of speech model outputs and language model inputs, and PLM with character-based tokenization is underexplored. In this work, we conduct extensive studies on how PLMs with different tokenization strategies affect spoken language understanding task including spoken question answering (SQA) and speech translation (ST). We further extend the idea to create T5lephone(pronounced as telephone), a variant of T5 that is pretrained using phonemicized text. We initialize T5lephone with existing PLMs to pretrain it using relatively lightweight computational resources. We reached state-of-the-art on NMSQA, and the T5lephone model exceeds T5 with other types of units on end-to-end SQA and ST.