Vocabulary Speech Recognition via Telephone

Due to the need for a flexible vocabulary and the high cost of collecting speech corpora, vocabulary independence of speech recognition systems has grown to be a significant issue. We describe the steps required to reach the objective of vocabulary-independent speech recognition and discuss our experimental experience with telephone speech recognition. There were two sets of experiments: 1) Wcommand recognition, where we compared vocabulary-independent (VI) and vocabulary-dependent (VD) systems as well as phonetic and word-based systems, and 2) 42-city name recognition, where our vocabulary independent recognition performance (8.5% W. Err. ) was very close to the VD performance (8%) of the Oregon Graduate Institute (OGI), which was reported by OGI, but significantly better than the VI performance (18%). Finally, we can say that some progress has been made. We identify the areas for improvement that are likely to move us closer to the goal of vocabulary independence, but much work needs to be done.

鉴于对灵活词汇的需求以及收集语音语料库的高昂成本，语音识别系统在词汇独立性方面的需求日益凸显。本文阐述了实现词汇独立语音识别目标所需的具体步骤，并就我们在电话语音识别方面的实验经验进行了讨论。实验分为两组：一是Wcommand识别实验，其中我们对词汇独立性（VI）与词汇依赖性（VD）系统以及基于音素和基于词的系统进行了比较；二是42个城市名称识别实验，我们的词汇独立性识别性能（8.5%的误识率）与俄勒冈研究生学院（OGI）报告的VD性能（8%）非常接近，但显著优于VI性能（18%）。最终，我们可以得出一些进展的结论。我们确定了可能推动我们更接近词汇独立性目标的改进领域，但仍需开展大量工作。