Search

Superficially, read and spontaneous speech—the two main kinds of training data for automatic speech recognition—appear as complementary, but are equal: pairs of texts and acoustic signals. Yet, spontaneous speech is typically harder for recognition. This is usually explained by different kinds of variation and noise, but there is a more fundamental deviation at play: for read speech, the audio signal is produced by recitation of the given text, whereas in spontaneous speech, the text is transcribed from a given signal. In this review, we embrace this difference by presenting a first introduction of causal reasoning into automatic speech recognition, and describing causality as a tool to study speaking styles and training data. After breaking down the data generation processes of read and spontaneous speech and analysing the domain from a causal perspective, we highlight how data generation by annotation must affect the interpretation of inference and performance. Our work discusses how various results from the causality literature regarding the impact of the direction of data generation mechanisms on learning and prediction apply to speech data. Finally, we argue how a causal perspective can support the understanding of models in speech processing regarding their behaviour, capabilities, and limitations. [ABSTRACT FROM AUTHOR]

Previous research has shown that differences in the marking of prosodic boundaries exist not only between varieties of the same language, but also between speaking styles of the same variety. Our aim is to gain insights into how these two factors interact in the case of German and Austrian German read and conversational speech. Analysing four acoustic cues, representing durational and fundamental frequency (f0) measures, we observed that pause duration was the strongest cue to prosodic boundaries and that f0 reset was the weakest, in both varieties and across speaking styles. In read speech however, we noticed differences between the two varieties in the weighting given to the four cues. We then employed the four investigated acoustic cues for the annotation of the less-resourced variety, by examining ways of exploiting resources from the more well-resourced variety. Testing three automatic approaches for boundary detection, we obtained an overall high performance: 82.4% and 76.6% F-score for read and conversational speech, respectively. Moreover, our results showed that training on more data from the well-resourced variety did not outperform the same system trained on less data from the target variety. • Prosodic boundaries in German and Austrian German, read and conversational speech. • Pause is the strongest cue, f0 reset is the weakest. • Cues weighting is more similar in conversational speech between the two varieties. • Automatic detection systems using the cues performed well on both speaking styles. • Training in variety-mismatched condition was not better than matched condition. [ABSTRACT FROM AUTHOR]

• first analysis of prosodic configurations mapped to information structural categories in Egyptian Arabic. • statistical analysis of holistic f0 contours with GAMMs and analysis of the interaction of acoustic features across the utterance with FPCA. • results yield a tentative prosodic focus-background construction across speakers, which is not applied by all speakers and thus not fully conventionalized for this specific purpose. • high inter-speaker variability is found and speaker-specific strategies are identified through reconstructions for individual speakers using FPCA. This paper investigates the prosody of sentences elicited in three Information Structure (IS) conditions: all-new, theme-rheme and rhematic focus-background. The sentences were produced by 18 speakers of Egyptian Arabic (EA). This is the first quantitative study to provide a comprehensive analysis of holistic f0 contours (by means of GAMM) and configurations of f0, duration and intensity (by means of FPCA) associated with the three IS conditions, both across and within speakers. A significant difference between focus-background and the other information structure conditions was found, but also strong inter-speaker variation in terms of strategies and the degree to which these strategies were applied. The results suggest that post-focus register lowering and the duration of the stressed syllables of the focused and the utterance-final word are more consistent cues to focus than a higher peak of the focus accent. In addition, some independence of duration and intensity from f0 could be identified. These results thus support the assumption that, when focus is marked prosodically in EA, it is marked by prominence. Nevertheless, the fact that a considerable number of EA speakers did not apply prosodic marking and the fact that prosodic focus marking was gradient rather than categorical suggest that EA does not have a fully conventionalized prosodic focus construction. [ABSTRACT FROM AUTHOR]