Linguistics

The current project aims to demonstrate how speakers of Liverpool English (LE) adapt their spoken style in different environmental contexts, and how this variation can be modelled for the application to an Automatic Speech Recognition (ASR) system. In tandem with this, the project will also be providing a systematic overview of the dialect features of Liverpool English.
It is not uncommon for an ASR system to have a degraded level of performance because of ambient noise, such as background noise, or reverberation off microphones. The effects that background noise can have on spectral distribution falls into two categories: stationary and non-stationary. A noise is stationary if it is constant within the time domain, and non-stationary if they vary within the time frame, such as in transient sound events like speaker interference and background music (Zhang et al., 2018). However, minimising the effects of background noise, particularly for non-stationary sound sources is still an issue for many ASR systems (e.g. Barker et al., 2013). The effects of environmental variation is not only constrained to ASR performance, it has also been found that speakers alter their spoken style in response to environmental factors (Junqua, 1996). Acoustic signal and production variation due to noise in communication is known as the Lombard effect. Variation in the input due to the Lombard effect can often result in poorer performance of an ASR system (Boril, Grezl & Hansen, 2011). In addition to variation in the environment, ASR systems are faced with the additional task of speaker variation, such as accent and dialect. Liverpool English has been extensively research in a multitude of dialect surveys. Watson (2007) has provided an overview of the salient features associated with LE and it was noted that the consonants of LE are phonologically identical to standard British English (BE), however the variation lies at the allophonic level (Watson, 2007). For example, plosive [t] is often affected by the phonological process of debuccalization and can be produced as [h] because of tightly constrained phonological conditions (Watson, 2002).
The aim of the pilot study is to investigate the effects of various speaker contexts on spoken style to be modelled for an ASR system. 12 participants (6 male and 6 female), aged 18 years and over from Liverpool, the Wirral and Merseyside will be recruited on a voluntary basis to participate in semi- structured interviews. The set of 3 interviews will take place in 2 locations around the University of Liverpool campus: a café setting where there is a vast amount of environmental variation, such as multiple talker babble and machine noises in the background. The second location proposed is a sound proof laboratory which will act as a control environment. Participants will take part in both a face to face interview and a telephone interview to elicit different conversational styles, in both a café setting and laboratory setting to observe the effects of background noise on spoken style in the two mediums. The final interview set will consist of reading a key word list in both settings to act as a control. A demographic questionnaire will also allow us to determine any effects of sociolinguistic patterning for the participants. Key questions will include, location, level of education and age group. The above data will be recorded, and analysed and annotated using PRAAT to allow for the identification of dialect- specific features of LE.