arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2406.02563eess.AScs.CLcs.SD
收藏

一种最小化代价方法用于修复端到端ASR系统中分词器的词表大小

A cost minimization approach to fix the vocabulary size in a tokenizer for an End-to-End ASR system

TCS Research(TCS研究中心)

Sunil Kumar Kopparapu, Ashish Panda

AI总结:

本文针对端到端ASR系统中分词器词表大小缺乏明确依据的问题,提出将分词过程视为黑盒并构建代价函数以最小化代价,从而选择最优token数量,实验证明该方法在LibriSpeech 100小时数据集上有效提升了系统性能。

摘要:

与混合语音识别系统将token的使用限制为单音素、双音素或三音素不同,端到端ASR系统中token的选择来源于训练数据的文本语料库。Byte Pair Encoding (BPE) 和 WordPiece 等分词算法在识别语音识别系统整体训练过程中使用的token方面很受欢迎。像 ESPNet 这样的流行工具包为这些分词算法使用预定义的词表大小(token数量),但没有讨论词表大小是如何得出的。在本文中,我们构建了一个代价函数,假设分词过程是一个黑盒,以便选择可能最有利于构建端到端 ASR 的 token 数量。我们通过在 LibriSpeech 100 小时数据集上的实验表明,当仔细选择 token 数量时,端到端 ASR 系统的性能会提高。

英文摘要:

Unlike hybrid speech recognition systems where the use of tokens was restricted to phones, biphones or triphones the choice of tokens in the end-to-end ASR systems is derived from the text corpus of the training data. The use of tokenization algorithms like Byte Pair Encoding (BPE) and WordPiece is popular in identifying the tokens that are used in the overall training process of the speech recognition system. Popular toolkits, like ESPNet use a pre-defined vocabulary size (number of tokens) for these tokenization algorithms, but there is no discussion on how vocabulary size was derived. In this paper, we build a cost function, assuming the tokenization process to be a black-box to enable choosing the number of tokens which might most benefit building an end-to-end ASR. We show through experiments on LibriSpeech 100 hour set that the performance of an end-to-end ASR system improves when the number of tokens are chosen carefully.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑