Yang Zhang

Junrui Ni, Liming Wang, Yang Zhang et al.

Recent advancements in supervised automatic speech recognition (ASR) have achieved remarkable performance, largely due to the growing availability of large transcribed speech corpora. However, most languages lack sufficient paired speech and text data to effectively train these systems. In this article, we tackle the challenge of developing ASR systems without paired speech and text corpora by proposing the removal of reliance on a phoneme lexicon. We explore a new research direction: word-level unsupervised ASR, and experimentally demonstrate that an unsupervised speech recognizer can emerge from joint speech-to-speech and text-to-text masked token-infilling. Using a curated speech corpus containing a fixed number of English words, our system iteratively refines the word segmentation structure and achieves a word error rate of between 20-23%, depending on the vocabulary size, without parallel transcripts, oracle word boundaries, or a pronunciation lexicon. This innovative model surpasses the performance of previous unsupervised ASR models under the lexicon-free setting.

Zhiqiang Zhan, Zifeng Hou, Yang Zhang

Recent neural models have shown significant progress in dialogue generation. Most generation models are based on language models. However, due to the Long Tail Phenomenon in linguistics, the trained models tend to generate words that appear frequently in training datasets, leading to a monotonous issue. To address this issue, we analyze a large corpus from Wikipedia and propose three frequency-based data normalization methods. We conduct extensive experiments based on transformers and three datasets respectively collected from social media, subtitles, and the industrial application. Experimental results demonstrate significant improvements in diversity and informativeness (defined as the numbers of nouns and verbs) of generated responses. More specifically, the unigram and bigram diversity are increased by 2.6%-12.6% and 2.2%-18.9% on the three datasets, respectively. Moreover, the informativeness, i.e. the numbers of nouns and verbs, are increased by 4.0%-7.0% and 1.4%-12.1%, respectively. Additionally, the simplicity and effectiveness enable our methods to be adapted to different generation models without much extra computational cost.