An Automatic System for Detecting Prosodic Prominence in American English Continuous Speech

Abstract

Abstract. A precise identification of prosodic phenomena and the construction of tools able to properly manage such phenomena are essential steps to disambiguate the meaning of certain utterances. In particular they are useful for a wide variety of tasks: automatic recognition of spontaneous speech, automatic enhancement of speechgeneration systems, solving ambiguities in natural language interpretation, the construction of large annotated language resources, such as prosodically tagged speech corpora, and teaching languages to foreign students using Computer Aided Language Learning (CALL) systems. This paper presents a study on the automatic detection of prosodic prominence in continuous speech, with particular reference to American English, but with good prospects of application to other languages. Prosodic prominence involves two different prosodic features: pitch accent and stress accent. Pitch accent is acoustically connected with fundamental frequency (F0) movements and overall syllable energy, whereas stress exhibits a strong correlation with syllable nuclei duration and mid-to-high-frequency emphasis. This paper shows that a careful measurement of these acoustic parameters, as well as the identification of their connection to prosodic parameters, makes it possible to build an automatic system capable of identifying prominent syllables in utterances with performance comparable with the inter-human agreement reported in the literature. Two different prominence detectors were studied and developed: the first uses a training corpus to set up thresholds properly, while the second uses a pure unsupervised method. In both cases, it is worth stressing that only acoustic parameters derived directly from speech waveforms are exploited.

Analyzing the Contribution of Top-Down Lexical and Bottom-Up Acoustic Cues in the Detection of Sentence Prominence

“…However, the combination of lexical and acoustic information adds only a small increase in the overall system performance, leading to an accuracy of 87%. Note that the performance of the unsupervised system is significantly higher than that of other unsupervised methods reported in the literature (ACC = 78.1% in [18] and 80.61% in [37]) using different feature combinations. However, direct comparison is not possible due to the different corpora used in different studies.…”

The Prosody of the Czech Discourse Marker ‘Jasně': An Analysis of Forms and Functions

“…Second, it is reasonable to assume that our rich and diverse semantic-pragmatic contexts will elicit a range of different pitch accents with different communicative functions, just as in other previous studies (Kohler, 2006:144;Niebuhr, 2010Niebuhr, , 2015; and if these pitch accents do differ from those realized in actual spontaneous speech, then it is further reasonable to assume that these differences will parallel those reported for the degree of assimilation, the weighting of F0 and duration in syllable prominence, or the range and slope of phrase-final intonation movements (Dilley and Pitt, 2007;Tamburini and Caini, 2005;Wagner et al, 2015). That is, we expect the differences to be quantitative but not qualitative.…”

Fundamental Frequency Height as a Resource for the Management of Overlap in Talk-in-Interaction

“…Detecting prominence automatically, particularly in spontaneous multi-party speech, is still a challenging research problem (e.g. Tamburini and Caini, 2005;Wang and Narayanan, 2006) although the composite measure used by Calhoun (this volume) offers a possible solution. Therefore we adopted the approach used in a number of quantitatively oriented studies of overlapping speech and prosody of discourse (Koiso et al, 1998;Shriberg et al, 2001b;Caspers, 2003) that segment the data into units delimited by short pauses.…”