Objective Determination of Speech Perception in Noise

• Applicant: Professor Dr.-Ing. Ulrich Hoppe
• Subject Area: Acoustics
• Term: from 2018 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 419293003

Speech perception is the most important social task of the auditory system. The audiometric determination of speech intelligibility can be performed with precise audiological test batteries. In particular, the Oldenburg Sentence Tests (OlSa) allows the highly reliable determination of speech intelligibility in noise. Unfortunately, these tests require a high-level of cooperation of the test persons. The long-term objective of the proposed project is the determination of speech intelligibility based upon Electroenecephalography (EEG). In particular, EEG measurements during speech perception will be analyzed in order to estimate the amount of speech perception. The project is based upon the hypothesis, that specific patterns in the EEG are changed by spectrotemporal speech features. Hence, speech intelligibility can be predicted by the EEG signal.During the last years several algorithms were developed which allow the determination of auditory attention. E.g. it is possible to use EEG for decision which one of two talkers is attended by a test person. These algorithms are manly based on correlation analysis of the speech envelope with the low-pass filtered EEG signal. All of them have two main disadvantages: 1. They make only use of the broadband envelope. Envelope changes in specific narrow bands remain undetected, 2. None of the algorithms incorporates auditory processing of neither outer, middle, or inner ear. Hence, they do not have a neural representation of the speech signal. By modifying the state of the art processing schemes and incorporating auditory processing we expect important insights in EEG-relevant speech processing.In a first step we calculate neural excitation patterns in response to speech for a set of Oldenburg sentences using a computational model of signal processing (Kates-Model). The most EEG-relevant speech signals will be extracted. Since the model delivers narrow-band excitation we can extract narrow-band envelopes. This assures the incorporation of narrow-band changes in speech signals. As a result, we expect a mapping function of speech signals on speech specific EEG. Since the EEG is affected not only by speech processing but other activities we have to consider a large group of subjects.In the second step, noise will be added to speech signals. Signal to noise ratios will be varied from nearly “non-intelligibility” to “easy-to-understand”. In a group of normal hearing adults we measure speech intelligibility and simultaneously the EEG. By analyzing these measurements we will be able to extend the model to speech in noise.As long-term, perspective we will be able to estimate speech intelligibility for individuals. As six-year perspective we plan to further develop the method to hearing impaired with known deficits.

• DFG Programme: Research Grants
• International Connection: United Kingdom
• Cooperation Partner: Professor Dr.-Ing. Stephan Weiss