Radek Skarnitzl

Many phonetic “truths” are based on descriptions of controlled speech material, and verifying their validity in spontaneous productions is essential. The present study investigates vowel duration as an acoustic correlate of stress in spontaneous English, in communicatively motivated contexts. By analyzing British and American political debates, this study aims to verify previously reported tendencies – specifically, that stressed vowels are significantly longer than unstressed ones. Our analysis of 16 speakers, based on linear mixed effects models, confirms the significant effect of stress on vowel duration and also considers additional factors influencing segmental duration like vowel length, phrase-final position, vowel height, or the nature of the following segment. In addition to stress, multiple regression analysis identifies vowel length, phrase-final position and vowel height as the most influential vowel duration predictors. Despite the variability of spontaneous speech, vowel duration proves to be a reliable correlate of stress, supporting the findings from controlled-speech research.

The pronunciation of the Spanish phonemes /b d ɡ/ is challenging for L2 learners, but most studies have relied on native speakers of English. This study examines the pronunciation of stop and approximant variants of Spanish /b d ɡ/ in two groups of Czech learners, 9 more and 9 less experienced users of Spanish, and compares it with the Spanish pronunciation norm as well as a group of 9 native speakers of Spanish. In total, 3280 items were coded as stop or approximant; alveolar productions of Spanish dental /d/ and labiodental productions of spelled were also noted. Logistic regression was used to predict the likelihood of a correct realization, and exploratory analyses were used for finer articulation place distinctions. The Spanish dental /d/ was revealed as most difficult for the Czech learners. A clear effect of spelling was shown for . Interesting cases of connected speech processes were also documented.

Lexical stress has received a lot of attention in Czech, but some questions remain unresolved. This paper examines two such aspects related to untypical stressing in Czech. The first concerns the length of anacruses (most sources admit the possibility of only one unstressed syllable at the beginning of a prosodic phrase). The second aspect is monosyllabic stress groups, and specifically situations which involve stress clash. Both anacruses and stress clash contexts are investigated in a corpus of two communicatively valid genres, newsreading and audiobooks (with 16 speakers for each genre), and related to the information structure of the utterances. The analysis revealed that more than one quarter of anacruses contain two (and rarely even three) syllables; typically such longer anacruses feature words which are contextually predictable and/or are semantically bleached. As for monosyllabic stress groups, more than two thirds have been surprisingly found to be formed by synsemantic rather than autosemantic words; these items thus reveal a prosodic-semantic conflict, with prosodic prominence occurring in informationally weak contexts.

Prosodic aspects of speech are crucial for the comprehensibility of L2 speakers, but prosody is rarely targeted in English language lessons. This paper describes an innovative training of English phrasal prosody using participants’ own speech as models in a modified listen-and-repeat paradigm, with their melodic and rhythmic patterns manipulated by means of PSOLA. The two-hour training was delivered individually to twelve intermediate native speakers of Czech. The comparison of a baseline recording of a read text before the training and two texts read six weeks after the training shows that using one’s own PSOLA-modified voice for prosody training is beneficial: the participants were perceived as sounding significantly more competent in the after-training recordings, their phrasing corresponded more to text-based predictions, and their melodic variability was significantly greater. The contribution of targeted prosody modifications in the teaching of L2 pronunciation are discussed.

Individual speakers’ voices display various unique patterns, one of the most prominent of which is vowel articulation. This study focuses on vowel space properties of 15 Czech speakers in read and spontaneous speech, comparing outputs of three formant extraction methods, measuring formants: (1) in the vowels’ temporal midpoints, (2) as their mean from the vowels’ middle thirds, and (3) in the vowels’ articulatory targets. The results show extensive variability across speakers, but also great within-speaker variability between the two speech styles, with spontaneous speech manifesting more centralised vowel pronunciation than read utterances. The first two measurement methods did not yield systematically different results, while formant values extracted from acoustically defined articulatory targets lead to noticeably larger vowel spaces. The results suggest that care should be taken when interpreting formant values obtained by different methods.

The so-called ‘mismatch’ is a factor which experts in the forensic voice comparison field encounter regularly. Therefore, we decided to explore to what extent the automatic-speaker-recognition system’s and the earwitness’ ability to identify speakers is influenced when recordings are acquired in different languages and at different times. 100 voices in a database of 300 recordings (100 speakers recorded in three mutually mismatched sessions) were compared with an automatic-speaker-recognition software VOCALISE based on i-vectors and x-vectors, and by 39 respondents in simulated voice parades. Both the automatic and perceptual approach seem to have yielded similar results in that the less complex the mismatch type, the more successful the identification. The results point to the superiority of the x-vector approach, and also to varying identification abilities of listeners.

Prosodic patterning is known to affect the impression that speakers make on their listeners. This study explores prosodic phrasing in good public speakers of American English and Czech. Czech is a West Slavic language whose intonation is reported to be flatter and prosodic phrases longer than in English. We analyzed prosodic characteristics of 10 speakers of Czech and American English who appeared in TED Talks, assuming such appearance to be a mark of a “good speaker.” Our objective was to see whether prosodic phrasing will be more similar in these public speeches between the two languages. We measured the length of prosodic phrases, speech rate in each phrase, and pitch range and melodic variability in the entire phrase, as well as in its pre-nuclear and nuclear portion. The number of syllables per phrase was higher in Czech than in English, although phrases were generally very short in both languages. The melodic indicators confirm smaller melodic variability in Czech even in publicly performed TED Talks. Overall, our results show that there are differences between Czech and English prosodic phrasing in good public speakers but that the genre also affects phrasing. Prosodic rendition—especially prosodic phrase length and melodic variability—is therefore a vital, albeit somewhat language-specific aspect of speech performance which public speakers should pay close attention to, both in their native language and in foreign languages.