Eliciting Meaningful Units from Speech

Abstract

Elicitation of information structure from speech is a crucial step in automatic speech understanding. In terms of both production and perception, we consider intonational phrase to be the basic meaningful unit of information structure in speech. The current paper presents a method of detecting these units in speech by processing both the recorded speech and its textual representation. Using syntactic information, we split text into small groups of words closely connected with each other. Assuming that intonational phrases are built from these small groups, we use acoustic information to reveal their actual boundaries. The procedure was initially developed for processing Russian speech, and we have achieved the best published results for this language with F1 equal to 0.91. We assume that it may be adapted for other languages that have some amount of read speech resources, including under-resourced languages. For comparison we have evaluated it on English material (Boston University Radio Speech Corpus). Our results, F1 of 0.76, are comparable with the top systems designed for English.

Automatic detection of prosodic boundaries in spontaneous speech

“…Existing automated phrase boundary detection methods often utilize lexical and syntactic cues along with acoustic input (e.g., [38][39][40]. They usually involve extensive preparation steps such as manual tagging (e.g., [41,42]) and training a specific, designated model (e.g., [38,39,41,43,44]).…”

“…They usually involve extensive preparation steps such as manual tagging (e.g., [41,42]) and training a specific, designated model (e.g., [38,39,41,43,44]). Approaches to speech segmentation based on acoustic signals alone were proposed in [45,46,40,47]. These efforts have been commonly applied to scripted speech (e.g.…”