On the End-to-End Solution to Mandarin-English Code-switching Speech Recognition
This work addresses speech recognition for bilingual speakers in conversational settings, representing an incremental improvement in a domain-specific area.
The paper tackles Mandarin-English code-switching speech recognition by proposing a multitask learning approach with language identification and vocabulary expansion, achieving improved performance on the SEAME corpus.
Code-switching (CS) refers to a linguistic phenomenon where a speaker uses different languages in an utterance or between alternating utterances. In this work, we study end-to-end (E2E) approaches to the Mandarin-English code-switching speech recognition (CSSR) task. We first examine the effectiveness of using data augmentation and byte-pair encoding (BPE) subword units. More importantly, we propose a multitask learning recipe, where a language identification task is explicitly learned in addition to the E2E speech recognition task. Furthermore, we introduce an efficient word vocabulary expansion method for language modeling to alleviate data sparsity issues under the code-switching scenario. Experimental results on the SEAME data, a Mandarin-English CS corpus, demonstrate the effectiveness of the proposed methods.