9-9.45 “Speech adaptations and intelligibility in challenging listening conditions: effects of aging.” Outi Tuomainen (UCL, UK)
Many older adults report having difficulties communicating in more demanding every-day situations such as when there are multiple talkers present in the background. Most research on age-related changes in speech communication in these complex listening environments has focused on problems concerning understanding speech in noisy environment and less attention has been given to whether these factors affect the production of speech. One key issue is whether older adults are as able as young adults to make ‘clear speech adaptations’ in challenging conditions in order to overcome communication difficulties. In this talk, I will first review age-related changes in speech production and perception and how these changes might impact our ability to communicate efficiently in background noise. I will then present data from our recent project where we investigated i) the acoustic characteristics of speech produced during interactions between pairs of young and older talkers in good and challenging listening conditions; ii) the intelligibility of these interactions to young and older listeners; and finally iii) what talker- and listener-related factors contribute to perceived intelligibility of spontaneous speech.
9.45-10.30 “Near-end Listening Enhancement: Overview & Recent Advances”, Henning Schepker (University of Oldenburg, Germany)
In typical speech communication scenarios speech is degraded by additive noise and/or reverberation, often leading to a reduced speech understanding. When the speech signal is available before playback, e.g., in public address systems or mobile telephony, so-called near-end listening enhancement algorithms can be applied that aim to improve speech understanding by appropriately preprocessing the speech signal. This talk is structured into two parts. In the first part we will provide a comprehensive overview on the problem of near-end listening enhancement and a review on the existing literature. In the second part we focus on a noise-adaptive near-end listening enhancement algorithm that combines time- and frequency-dependent amplification and dynamic range compression. We present extensions for noisy and reverberant scenarios as well as scenarios in which only a noisy version of the clean speech is available. Furthermore, we show that ratings of perceived listening effort can be used to assess the benefit of near-end listening enhancement algorithms in scenarios where speech intelligibility is already perfect.
11-11.45 “Speech modifications inspired by natural speech”, Maria Koutsogiannaki (BCBL, Spain).
In this talk we will focus on speech modifications for enhancing the intelligibility of speech in noise. We will present two families of speech intelligibility enhancement algorithms: SNR-based algorithms and modifications independent of the type and the level of noise. The latter exploit acoustic properties of elicited natural speech of low and high intelligibility. Therefore, in order to perform modifications as humans do when they speak in dynamically-changing listening environments, we will present the acoustic differences between these speaking styles and the signal processing theory of how to estimate these acoustic features. During the hands on session we will put the theory into practice by performing analysis and modifications.
11.45-12.30 “Speech production in noise: Lombard effect, speech intelligibility and listening difficulty”, Simone Graetzer (University of Liverpool, UK)
Speech produced in a noisy environment differs temporally and spectrally from speech produced in quiet conditions. These differences, which include an increase in the speech level with the level of the noise, are known collectively as the Lombard effect. When transmitted at the same signal-to-noise ratio, “Lombard speech” is more intelligible than speech produced in quiet. Therefore, the changes that occur in Lombard speech are of interest for speech enhancement. In addition, they are of interest to developers of speech recognition systems, which should function robustly in adverse noise scenarios. In this talk, a review of the literature concerning speech production in noise will be provided. There will be a discussion of the effects of such noise on speech intelligibility and listening difficulty, and, briefly, of how strategies relying on insights drawn from Lombard speech have been used in recent years for speech enhancement.