Removing speech artifacts from electroencephalographic recordings during overt picture naming

Abstract

A number of electroencephalography (EEG) studies have investigated the time course of brain activation during overt word production. The interpretation of their results is complicated by the fact that articulatory movements may mask the cognitive components of interest. The first aim of the present study was to investigate when speech artifacts occur during word production planning and what effects they have on the spatio-temporal neural activation pattern. The second aim was to propose a new method that strongly attenuates speech artifacts during overt picture naming and to compare it with existing methods. EEG and surface electromyograms (EMGs) of the lips were recorded while participants overtly named pictures in a picture-word interference paradigm. The comparison of the raw data with lip EMG and the comparison of source localizations of raw and corrected EEG data showed that speech artifacts occurred mainly from ~400 ms post-stimulus onset, but some earlier artifacts mean that they occur much earlier than hitherto assumed. We compared previously used methods of speech artifacts removal (SAR) with a new method, which is based on Independent Component Analysis (SAR-ICA). Our new method clearly outperformed other methods. In contrast to other methods, there was only a weak correlation between the lip EMG and the corrected data by SAR-ICA. Also, only the data corrected with our method showed activation of cerebral sources consistent with meta-analyses of word production.

Neural correlates of spoken word production in semantic and phonological blocked cyclic naming

“…Given that the positive component in the posterior region is relatively late (which peaked around 450 ms), a cautionary note is that it may be subjected to the influence of speech-related artifacts. Although low-pass filtering and artifact rejection is expected to remove any speech-related artifacts, other solutions for speechrelated artifact rejection have also been proposed such as the SAR-ICA procedure in Porcaro, Medaglia, and Krott (2015). Thus, replication of the current finding, preferably with more than one artifact rejection methods, would be important for future research.…”

On the cortical dynamics of word production: a review of the MEG evidence

“…It is also potentially possible to use other component separation techniques to separate noise from signal, and spatial filtering techniques are implemented in current freely available electrophysiological analysis packages, such as Brainstorm (Tadel, Baillet, Mosher, Pantazis, & Leahy., 2011) and Fieldtrip (Oostenveld, Fries, Maris, & Schoffelen, 2011) to enable trained users to identify and remove EKG, blink, and other sources of artefactual noise. In EEG, the use of blind source separation based on the canonical correlation analysis and the independent component analysis has been employed and evaluated with interesting results (De Vos et al, 2010;McMenamin et al, 2010;Porcaro, Medaglia & Krott, 2015); to our knowledge, no such evaluation has yet been conducted using MEG data. There is also the consideration that the nature of MEG signal allows for good separation of muscular and cortical sources on spatial grounds when compared to EEG: the beamforming analysis can allow muscular noise to be separated from cortical signal if allowed to project sources in the entire head space instead of constraining them to be inside the pial surface (see Laaksonen et al, 2012, Supplementary Figure 2B and 2C for a demonstration of this).…”

Primary phonological planning units in spoken word production are language-specific: Evidence from an ERP study

“…Although recent studies have started using the ERP technique to investigate the early processes in overt speech production 31–34 , these early brain responses are not necessarily free of artifacts. In an overt naming task with 850-ms average naming latency, Porcaro, Medaglia and Krott 35 adopted an Independent Component Analysis procedure to remove articulation-related artifacts, and identified a major artifact after 400 ms post picture onset as well as a smaller but earlier artifact around 160 ms. To minimize the possible artifacts in the early processes, we chose to use a delayed naming task in the current study. Pictures stayed on the screen for 800 ms followed by a question mark (i.e., the cue), and participants were required to prepare the name of the picture as soon as possible but to withhold their naming response until the cue onset.…”