EFR Analysis

The analysis steps described below were completed in MATLAB (MathWorks, MA, USA) with and without bandpass filtering (BPF) the EEG between 65 and 130 Hz using a 12^th order infinite impulse response Butterworth design and MATLAB’s zero-phase filtfilt function. Each stimulus trial was divided into 4 epochs of ~ 1 s for the purpose of noise rejection. A noise metric was calculated as the average EEG amplitude between 80 and 240 Hz when no bandpass filtering (no-BPF) was used (Easwar et al. 2020, 2021a, b), and between 65 and 130 Hz when bandpass filtering was used (with-BPF). Different frequency ranges were used to correspond with the filter conditions during EFR estimation. A noise rejection threshold was computed in each participant, and this varied for each filtering condition. The threshold was equal to the third quartile + 1.5 times the inter-quartile range. Epochs that exceeded the computed threshold were discarded prior to averaging.

EEG was averaged over the two stimulus polarities, and EFR amplitude was estimated using a FA (Aiken and Picton 2006; Choi et al. 2013). For all three vowels, the analysis window was 350 ms and began 17 ms after vowel onset. For the FA, the f₀ time course for each vowel was determined using Praat. For each f₀ time course, reference cosine and sine sinusoids were created based on the f₀ frequency. To align the f₀ time course with the EEG, processing delays ranging from 5 to 25 ms in 0.5 ms increments were used (total n = 41). The processing delays shifted the EEG to an earlier time to ‘un-do’ the cochlear and neurophysiologic processing delay. The range was chosen based on estimated EFR latencies for modulation frequencies/f₀ between 80 and 190 Hz (Cohen et al. 1991; John and Picton 2000; Picton et al. 2003; Purcell et al. 2004; Alaerts et al. 2010; Bidelman 2018; Canneyt et al. 2021). At each assumed processing delay, the time-shifted EEG was multiplied with the reference sinusoids to obtain real and imaginary components of the EFR. The real and the imaginary components were independently averaged over the entire analysis window, and subsequently combined in a complex number to estimate EFR amplitude and phase. Residual noise for both response fundamental frequencies was computed as the average EEG amplitude in 6 bins below and 8 bins above the two response fundamental frequencies, excluding the bins between response fundamental frequencies (Easwar et al. 2015a). The effective frequency resolution of the Fourier analyzer was 3.46 Hz.