Time is of the essence: an evaluation of temporal compression algorithms

Although speech is a potentially rich information source, a major barrier to exploiting speech archives is the lack of useful tools for efficiently accessing lengthy speech recordings. This paper develops and evaluates techniques for temporal compression - reducing the time people take to listen to a recording while still extracting critical information. We first describe an exploratory study that identifies novel excision techniques that remove unimportant words or utterances from the recording. We then develop a new method for evaluating how well temporal compression supports users in forming a general understanding of a recording. Applying this method, we demonstrate that excision techniques are generally more effective than standard compression techniques that simply speed up the entire recording.