语音识别视角下四川方言数字语音特征分析

An Analysis of Sichuan Dialect Digit Speech Features from the Perspective of Speech Recognition

摘要: 方言数字识别是当前语音识别的难点。本文以四川方言为例，就方言数字的语音特征开展分析，指出方言语音的地域和社会分层差异所致的语音变体，以及连续数字中韵律单位的基本分布情况，并就3000余条ASR数字识别结果进行了实证分析。本文认为，“左重”韵律模式对四川连续数字的识别产生了重要的影响。

Abstract: Dialect digit recognition remains a challenging frontier in the field of automatic speech recognition (ASR). This paper takes Sichuan dialect as a case study to analyze the phonetic characteristics of dialectal numerals. It highlights the speech variations caused by regional and social stratification differences within the dialect, outlines the fundamental distribution of prosodic units in continuous digit sequences, and presents an empirical analysis based on over 3000 ASR digit recognition results. The study concludes that the “left-heavy” (LH) rhythmic pattern exerts a significant influence on the recognition accuracy of continuous Sichuan dialect digits.