A cost minimization approach to fix the vocabulary size in a tokenizer for an End-to-End ASR system
This addresses a specific, incremental issue for ASR researchers by providing a method to optimize tokenizer vocabulary size, which is often arbitrarily set in existing toolkits.
The paper tackles the problem of determining an optimal vocabulary size for tokenizers in end-to-end ASR systems, proposing a cost minimization approach that improves performance, as shown by experiments on the LibriSpeech 100-hour dataset.
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.