Materiały konferencyjne (WA)

Permanent URI for this collection

https://repozytorium.amu.edu.pl/handle/10593/3290

Browse

Now showing 1 - 5 of 5

Automatic English phoneme recognition from articulatory data generated by EPG systems with grid and anatomical layout of contact sensors
(Australasian Speech Science and Technology Association Inc., 2019-08-22) Krynicki, Grzegorz; Dziubalska-Kołaczyk, Katarzyna; Weckwerth, Jarosław; Michalski, Grzegorz; Kaźmierski, Kamil; Maciejewska, Barbara; Wiskirska-Woźnica, Bożena; Żygis, Marzena; Kuczko, Wiesław; Sekuła, Alicja
The aim of the study was to conduct automatic phoneme identification from articulatory data that accompanied the production of these phonemes in continuous speech. The articulatory data were obtained from 2 electropalatographic systems, Palatometer by Complete Speech and Linguagraph by Rose-Medical. Palatometer was used with the artificial palate containing 124 contact sensors in a grid layout, including 2 sensors monitoring the lip contact. The palate included a vacuum-thermoformed flexible printed circuit. Linguagraph was used with the acrylic artificial palate designed and developed for the purpose of this study, containing 62 electrodes in anatomical layout. Palatometer was used by one native of General American and Linguagraph by one native of General British, each reading 140 phonetically balanced sentences that included Harvard Sentences and TIMIT prompts. The EPG data were parametrised into dimensionality reduction indexes, which were analysed by means of linear discriminant analysis and a probabilistic neural network. The results of classifications are discussed.
Comparing rhythm in speech and music: the case of English and Polish
(2014-09-27) Jekiel, Mateusz
Although linguistic and musical rhythm have been widely studied by linguists and musicologists alike, there is still a scarcity of quantitative studies that would examine the extent of the relationship between those two domains. A study by Patel et al. (2006) suggests that some characteristic features of a culture’s language are reflected in its instrumental music. The method applied in the study used the normalised pairwise variability index (nPVI), a measure of temporal patterning in speech, introduced by Low (1998) and used primarily for comparing stresstimed and syllabletimed languages (Grabe and Low 2002). By comparing the variability of vocalic duration in recorded speech with nPVI values computed from music notation, Patel et al. concluded that the language rhythm of English and French is mirrored in the music of corresponding English and French classical composers of the 19th c. Although this interdisciplinary approach has been investigated in recent studies (e.g. McGowan and Levitt 2011), it still brings more questions than answers. Firstly, it is difficult to ascertain whether the method can be applicable to all musical forms, as classical music from the 19th c., a period identified as the age of musical nationalism, limits the scope of the studies. Secondly, relying solely on musical notation means omitting recorded live performances that might carry different rhythmic information. Finally, while English and French represent a stressedtime and a syllabletimed language respectively, the differences found in the study by Patel et al. might not be so evident for other languages or dialects. The aim of this study is to attest the method used in Patel et al. (2006) by examining a different set of data, focusing on English and Polish rhythm in speech and music. The speech corpus consisted of 20 English and Polish recorded sentences. The music corpus was divided into two categories and consisted of recordings of live performances and musical notation: (1) the classical music category comprised of a selection of 19th c. themes of English and Polish composers (e.g. Elgar’s and Chopin’s), while (2) the folk music category comprised of a set of traditional English and Polish folk songs (Luboff and Stracke 1969). It can be predicted that there will be a similar discrepancy between the nPVI values for Polish and English language and musical rhythm as in the study by Patel et al.. However, the interrelationship between those results is difficult to foresee and might vary from the original experiment. The outcomes of this study will hopefully shed more light on the relationship between language rhythm and musical rhythm and open new paths for future interdisciplinary studies.
Cross-dialectal analysis of English pitch register and its influence on perceived speech friendliness
(2015-02-23) Malarski, Kamil; Jekiel, Mateusz
Intonation is a pragmatically meaningful cue in evaluating speech (Grabe et al. 2003). While intonational contours may be interpreted differently depending on the context, pitch range has been named a speech parameter especially conducive to judging speaker’s friendliness. As Ohala (1983) explains, pitch range can be analysed either as pitch span, which is the fluctuation between the lowest and the highest pitch level in the speaker’s voice, or as pitch register, i.e. the average vocal frequency of an individual speaker. In our previous research, a comparison of pitch register in Dutch, English and Polish led us to the conclusion that languages with a higher pitch register are perceived as more friendly. Moreover, we noted that even a slight difference of 5 Hz in frequency is noticeable by listeners and affects their judgments. Similarly, differences in pitch range are observable across different varieties of English and can play a role in their reception (Cruttenden 1994: 141, Malarski 2013). Therefore, the present study focuses on these differences across several dialects of English with the aim to investigate the potential effect of pitch range on perceived speech friendliness. For the present study, we recorded two male middle-aged speakers for each of the following accents of English: Southern British, Manchester, Australian, General American and Canadian. We selected short fragments of spontaneous speech and modified intonation by lowering and raising pitch register in the recordings by 5Hz and 10Hz using Praat (Boersma and Weenink 2013). The material was randomized and prepared for a listening survey. 50 students at Adam Mickiewicz University in Poznań were asked to evaluate the recordings in terms of friendliness on a 7-point Likert scale. The additional criteria were attractiveness, self-confidence and prestige. Our predictions are that students will 1) rate the accents with higher pitch register as more friendly, 2) rate modified recordings with raised pitch as more friendly, and 3) rate modified recordings with lowered pitch as less friendly. If these hypotheses are confirmed, they will constitute new evidence for stating that suprasegmentals, and pitch register in particular, are powerful cues for listeners in pragmatic judgments of speech.
Cross-linguistic analysis of pitch range and its influence on perceived speech friendliness
(2015-02-20) Jekiel, Mateusz; Malarski, Kamil
Intonation is an important cue in evaluating speech of other speakers. Studies have shown how pragmatically meaningful it is (Grabe et al. 2003). Apart from intonational contours that may be interpreted differently depending on the context, pitch range has been named a speech parameter that is especially conducive to judging whether a speaker sounds friendly or unfriendly. Chen, Rietveld and Gussenhoven (2001) compared pitch range in Dutch and English, concluding that the latter is perceived as more friendly than the former due to its larger pitch range. This conclusion has led us to investigate the status of Polish in this analysis. In the present study, we recorded two adult male native speakers of Dutch, English and Polish and compared their pitch span and register (Ohala 1983). English was found to have a wider pitch range than Dutch, similarly to the previous study, while Polish was found to have a wider pitch range than Dutch but narrower than English. In the second part of the experiment, our subjects will evaluate how friendly the speech recordings will sound to them. Apart from raw speech samples, we have prepared modified recordings using Praat (Boersma and Weenink 2013), where both pitch span and register were set at the same levels across all speakers. We have prepared an online survey, comprising of a set of randomly ordered recordings of Dutch, English and Polish declarative, lexically identical sentences, both raw and acoustically modified. The participants (at least 20 native speakers of Dutch, English and Polish) will assess the speakers in terms of friendliness using a sevenpoint Likert scale. The survey will be preceded by a questionnaire concerning participants’ L1 and L2, age, gender, etc. Our hypotheses are that 1) the scores for the modified recordings will differ from the unmodified ones, signifying that pitch range is the main criterion for judging whether a speaker sounds friendly or unfriendly, 2) Dutch will be perceived as the least friendly due to its narrow pitch range, 3) Polish will be perceived as less friendly than English. Polish has rarely been featured in language attitude studies so far, hence, it is especially interesting to see how it scores in relation to English and Dutch. If the intermediate status of Polish pitch patterns is confirmed, yet another evidence will be obtained in the support of the view that pitch is the primary cue for determining speech friendliness.
Dynamic information for Polish and English vowels in syllable onsets and offsets
(2014-09-27) Jekiel, Mateusz
According to the simple target model, vowel targets can be described as “a unifying concept among articulatory, acoustic and perceptual characterizations of vowels” (Strange 1989). Moreover, the model states that the first two oral formants (F1/F2) give sufficient information for vowel identification and the problems with vowel perception derive from variations in their production. Although there is a great deal of acoustic variability for a speaker producing a vowel in different consonantal contexts, the relationship between the vowel targets and the produced vowels still remains, as the listeners can recognise the intended vowels despite the variations. A study by Lindblom and Studdert-Kennedy (1967) argues that listeners compensate for the production undershoot by a perceptual overshoot. Further research in the dynamic specification model proved that vowels in consonant-vowel-consonant (CVC) contexts are identified more accurately then in isolation (Verbrugge et al. 1976), syllable onsets and offsets carry useful information for vowel identification (Strange et at. 1983) and a combination of both syllable onsets and offsets provides a “complex dynamic signal that is highly informative” for vowel identification (Jenkins and Strange 1999: 1208). As this paper deals with Polish learners of English, another key study is by Iverson and Evans (2007), showing that L2 learners with a simple vowel system are less successful in identifying English vowels and use different cues than learners with a complex vowel system. The experiment, based on the paper by Jenkins and Strange (1999), was carried out at the School of English at Adam Mickiewicz University in Poznan. A set of Polish and English vowels in CVC contexts was recorded and modified in Praat according to five different conditions: silent-centre (three initial pitch periods, four final pitch periods), one initial pitch period, five initial pitch periods, four final pitch periods and eight final pitch periods, attenuating the remaining vowel components to silence. The participants of the experiment, five students of AMU School of English (proficiency level) and five students from non-linguistic studies (intermediate level), were asked to identify the vowel they heard by marking the key word on an answer sheet that contained the same vowel. The purpose of the experiment was to establish the adequate amount of dynamic information in syllable onsets and offsets needed for correct vowel identification, to compare Polish and English vowel perception and to see to what extent the dynamic specification model describes Polish speakers' vowel perception. The results showed that (1) both groups had difficulties in identifying English vowels, (2) proficiency-level students had better results from their intermediate-level counterparts (80% in silent-centre condition vs. 48%) and (3) both groups had exceptional results in identifying Polish vowels (over 90% in all conditions). Firstly, the difference between a simple L1 (Polish) and a complex L2 (English) vowel system seems to have a bearing on correct vowel identification. Secondly, it seems possible for EFL students to use the dynamic information for English vowels, as the group’s results correlated with the one’s from Jenkins and Strange (1999). Lastly, the results show a difference between Polish and English vowel perception, as the dynamic information in syllable onsets and offsets seems rather futile for the correct identification of Polish vowels.

Browse

Browsing Materiały konferencyjne (WA) by Subject "English"

Results Per Page

Sort Options