Towards an objective measure of speech understanding

Objectives: Speech understanding is normally measured with behavioural tests. Such tests require the active participation of the test subject and are strongly influenced by the knowledge of language and cognitive processes. We aim to develop an objective and automatic measure of speech understanding based on EEG measurements. The purpose of this study was to examine if acoustic signals with different SNRs elicit systematically different auditory responses and thus could be used to estimate the speech-reception threshold objectively. Methods: 5 normal-hearing adults participated in this study. The subjects listened attentively to two different speech materials embedded in speech weighted noise at three different SNRs (-2, -4 and -6 dB) while the EEG was being recorded. We determined the envelopes of the clean speech signals, and used them to calculate a decoder that estimates the stimulus envelope based on the EEG data. The decoder is based on the principle that the low frequencies (2-8 Hz) of the temporal envelope of the stimulus can be linearly related to the EEG. We expressed the correctness of the estimated envelope as a difference measure between the estimated envelope and the envelope of the clean speech signal. Results: In the high-SNR conditions the estimated envelope showed less difference with the envelope of the clean speech stimulus than in the low-SNR conditions, meaning that the stimulus was better represented in the EEG data and thus presumably better understood. Conclusion: Using EEG measurements we are able to estimate the SNR of speech signal in noise presented to the subject. This is the first step towards automatically and objectively estimating speech understanding.status: accepte