Automatic indexing of key sentences for lecture archives using statistics of presumed discourse markers

Abstract

Automatic extraction of key sentences from lecture audio archives is addressed. The method makes use of the characteristic expressions used in initial utterances of sections, which are defined as discourse markers and derived in a totally unsupervised manner based on word statistics. The statistics of the presumed discourse markers are then used to define the importance of the sentences. It is also combined with the conventional tf-idf measure of content words. Experimental results using a large corpus of lectures confirm the effectiveness of the method based on the discourse markers and its combination with the keyword-based method. It is also shown that the method is robust against ASR errors and sentence segmentation accuracy is more vital. Thus, we also enhance the segmentation by incorporating prosodic information.

A New ASR Evaluation Measure and Minimum Bayes-Risk Decoding for Open-domain Speech Understanding

“…Comparing the cases (2) and (3) in Table 2, it is observed that the ASR degraded accuracy, especially on the precision. In [10], however, we showed that major cause is incorrect sentence segmentation by automatic period insertion rather than word substitution errors.…”

“…It is equivalent to the orthodox speech recognition described in equation (9) when the 0/1 loss function is used in equation (10). In our baseline ASR system, this decoding is used.…”

Spoken language processing for audio archives of lectures and panel discussions

“…We have proposed a method to automatically train a set of discourse markers without any manual tags, and shown the effectiveness in segmentation of lecture audio [19,5]. Then, we apply the discourse segmentation to extraction of key sentences from lectures [20,5] for generating more informative tags of the indices.…”

Speech-based real-time presentation tracking using semantic matching

“…Previous studies either provide very limited support for deviations from presentation script [78] or only perform off-line alignment of speech and presentation content [61,88]. Real-time tracking of presentation content is a novel approach for providing content-based assistance during presentations.…”

“…Kawahara et al [78] extracted characteristic keywords of the lecture using tf-idf weighting and then used the extracted keywords as one of the measures for indexing key sentences in lecture archives.…”