Skip to content

Speech-induced Artifacts Can Contaminate Intracranial Recordings in Awake Patients

Key findings

  • This paper reports the identification and quantification of speech-induced artifacts in several types of intracranial recordings obtained using a speech production task during the implantation of deep brain stimulation electrodes
  • The artifacts were caused by mechanical vibrations, and they could also be found in a "blank" pin not connected to any electrode
  • Caution should be exercised when interpreting speech-evoked high-gamma activity and frequency following responses, as the artifact affects these signal
  • Intracranial recordings obtained during speech production tasks should be carefully assessed for speech-induced artifacts

Surgery to place deep brain stimulation (DBS) electrodes in awake patients has provided unique opportunities to study neural control of speech production. One pronounced advantage of intracranial recordings is that they can capture neural activity in the high-gamma frequency band (60–200 Hz), considered potentially useful for developing brain–computer interfaces for speech prostheses.

However, the quality of these recordings may be affected to various degrees by the same movements that sometimes contaminate noninvasive recordings, such as facial and mouth movement artifacts observed on scalp EEG.

Alan Bush, PhD, an instructor in the Department of Neurosurgery at Massachusetts General Hospital, Mark Richardson, MD, PhD, director of the Functional Neurosurgery Program, and colleagues have detected artifacts caused by speech-induced mechanical vibrations in the high gamma frequency band during DBS surgery. In NeuroImage, they suggest approaches to data collection and analysis that might reduce the potential for these false discoveries.


The study participants were 30 patients with Parkinson's disease who were undergoing awake neurosurgery for implantation of DBS microelectrodes in the subthalamic nucleus. Before recording started, one or two electrocorticography strips were placed subdurally through the standard burr hole, targeting the left superior temporal gyrus and left inferior frontal gyrus.

During surgery, participants repeated sequences of syllables presented to them through earphones. Each performed two or three sessions of the task with the microelectrodes positioned at different depths within the subthalamic nucleus. After the left DBS lead was implanted, some participants performed a final session of the speech task to provide data from DBS leads.

In most participants, one pin in the electrocorticography array was not in contact with the brain because of the configuration of the connectors used. The team recorded the signal from this "blank" pin anyway because it served as a control for non-neural sources of noise affecting the neural signal.

Key Results

The researchers identified a narrowband gamma component in the neural signals that was consistent with a mechanically induced artifact. The artifact was:

  • Widespread, as it affected recordings obtained from macro contacts on the shaft of the microelectrodes, from electrocorticography strips, from the blank pin not connected to any electrode, and from the implanted DBS leads
  • The most prominent feature in many electrode recordings
  • Detected almost exclusively during cued speech, not during listening
  • Not localized to the auditory cortex or any other cortical region

Clinical Implications

The frequency following response (FFR) is a brainstem response to auditory stimuli that can be recorded from scalp electrodes. Its detection has become a powerful diagnostic tool in audiology and neurology, known as the auditory brainstem response.

In light of these results and similar results recently published in the Journal of Neural Engineering, caution is necessary when interpreting high-gamma activity and FFR during speech production. In particular, the FFR signal is exactly what would be expected if there was a speech-induced vibration artifact.

Identifying and Removing the Artifact

Several methods might be able to identify speech-induced vibrations:

  • Techniques that correlate the audio signal with the neural data can detect the presence of acoustic contamination
  • Recording from blank pins provides a control for non-neural sources affecting the signal
  • Vibration sensors can detect mechanical vibrations along the recording system that might affect the signal

The magnitude of the artifact varied across subjects, across electrodes within the same subject, and between recording sessions for the same subject. The authors are currently creating data-driven tools for removing these artifacts from intracranial recordings, which will help advance the field of intracranial speech neuroscience.

Visit the Functional Neurosurgery Program

Refer a patient to the Department of Neurosurgery

Related topics


Vasileios Kokkinos, PhD, and R. Mark Richardson, MD, PhD, of the Department of Neurosurgery, and colleagues have conducted the first direct testing that demonstrates hippocampal spindles and barques on intracranial EEG are normal findings.


Massachusetts General Hospital researchers modified a high-resolution neural recording system and developed innovative techniques for applying it to human research. Their advances lay the foundation for better understanding of neurologic diseases, brain cancer and how to enhance neurorestorative technologies.