The Centre for Speech Technology Research, The university of Edinburgh

PWorkshop Archives: Autumn Term 2003

14 Oct 2003

Bert Remijsen, Vincent van Heuven and Farienne Martis

Papiamentu prosody and the word-accent type

In the field of prosody, the word-accent type is well-known from Swedish and Serbocroatian. In these languages, the stressed syllable can have either of two tonal patterns, distinct from one another in terms of alignment. In this paper, we argue that such an analysis is also appropriate for Papiamentu, a Caribbean creole. In addition, however, Papiamentu word-accent interacts with an independent lexical stress feature. And also, the choice of word-accent is partly determined by morphosyntactic considerations.



28 Oct 2003

Paul Warren (Victoria University of Wellington, NZ)

We have fears to suit every traveller -- effects of sound change on word recognition In this talk I will present data from a collaborative project being carried out in Wellington and Christchurch that looks at the consequences for word recognition of a change-in-progress in the pronunciation of New Zealand English. The change is the on-going merger towards NEAR of the front-centering NEAR and SQUARE diphthongs. Our production data confirms that the merger is very advanced for young New Zealanders, though a comparison of data from the two centres shows a more complete merger for the Wellington group. The Christchurch group shows more variability in their realisations of the SQUARE vowel, but this variability is correlated with the same participants^? scores in an identification task for minimal pair words with these diphthongs. The Wellington group fails to distinguish the two vowels in production but scores extremely well in the identification task. Response times from this same group in an associative priming task indicates an asymmetry in lexical access processes that mirrors the asymmetry in the progress of the merger. Words with a NEAR vowel (^?cheer^?) prime associates of words with the SQUARE vowel (^?sit^? as an associate of ^?chair^?) as well as those of words with the NEAR vowel (^?shout^?), while words with a SQUARE vowel only prime their own associates (^?chair^? primes ^?sit^? but not ^?shout^?). I will discuss the implications of our sets of findings for issues in spoken word recognition, and outline some of the further questions raised in our research project.


04 Nov 2003

Jim Scobbie (QMUC)

Phonetics and phonology of English /l/ sandhi: ultrasound data


11 Nov 2003

Steve Renals

Speech and crosstalk detection in multi-channel audio

The analysis of scenarios in which a number of microphones record the activity of speakers, such as in a roundtable meeting, presents a number of computational challenges. For example, if each participant wears a microphone, it can receive speech from both the microphone's wearer (local speech) and other participants (crosstalk). The recorded audio can be broadly classified in four ways: local speech, crosstalk plus local speech, crosstalk alone and silence. In this talk I shall discuss some investigations related to the automatic classification of audio into these four classes. In particular, I shall discuss the utility of various acoustic features (eg kurtosis, cross-correlation metrics and fundamentalness) for this problem, and the construction of some simple statistical models that use these features.

(Joint work with Stuart Wrigley and Guy Brown.)



25 Nov 2003

Tim Mills and Hannele Nicholson

Cues to voicing contrasts in whispered Scottish obstruants

Previous studies have shown that contrary to expectations pitch is perceptible in whispered Chinese, Thai, Norwegian, and Swedish. An acoustic analysis of whispered Norwegian is presented which further previous perceptual results. This paper also shows from perceptual and acoustic data of Scottish English that the distinction between whispered voiced and voiceless consonants is perceptible as well. We conclude that, while differences in phonation may be significant cues to these contrasts, there are other cues that serve as adequate "backup", both in normal and whispered speech.



02 Dec 2003

No P-Workshop (Linguistics circle instead)



09 Dec 2003

Olga Gordeeva (Queen Margaret University College)

Dynamics of bilingual sound structure learning and language interaction.

I present intermediate results of instrumental acoustic analysis of the degree to which young bilinguals separate their language structures in the production of phonological encoding of tense and lax vowels, and the Scottish Vowel Length Rule. I collected longitudinal data for two Russian-Scottish English children, aged 3;4 to 4;8. Russian is spoken in their families, and Scottish English in the community (Edinburgh). The children differ in the amount of input in their languages. The results presented here cover the 1st and 2nd longitudinal moments (age 3;2 to 4;0).

I emphasise the importance of establishing an appropriate control framework for bilingual child speech. I show that (1) some apparent language interaction structures can in fact be explained by speech immaturity, and by language variability in the input (child-directed speech); (2) the dimension of language dominance needs to be involved to explain the opposite direction in which the language interaction takes place in the speech of our two subjects for the same competing cross-linguistic sound structures.



[back to PWorkshop Archives]

<owner-pworkshop@ling.ed.ac.uk>