Skip to content

← Publications

2012 preprint methods ArXiv.org

Algorithm to suppress scanner noise in recorded speech during functional magnetic resonance imaging

Satrajit S Ghosh

Identifiers and access

DOI
10.48550/arxiv.1207.5827
arXiv
1207.5827
PDF
Open-access copy →

Key findings

Presents a single-channel adaptive filter combining time- and frequency-domain elements to suppress repetitive fMRI scanner noise in recorded speech, achieving significant SNR gains suitable for both real-time patient-operator communication and offline speech analysis during image acquisition.

Abstract

Source: arxiv

The high-intensity, repetitive noise associated with functional magnetic resonance imaging hinders on-line monitoring of subjects' speech and/or recording speech signals suitable for off-line analysis. The proposed algorithm enhances the speech signal by suppressing the scanner noise in the signal recorded by a single-channel microphone. Significant increases in signal-to-noise ratio are achieved using an adaptive filter that combines time and frequency domain elements. In addition to providing a recording suitable for speech analysis, such a real-time system provides an alternative means (to, e.g., the "panic ball") for communication between the patient and the operator during image acquisition.

Topics

  • speech-voice-biomarkers
  • neuroimaging-methods

Lab authors

This record was curated from the lab's CV, NCBI MyBibliography, and OpenAlex. See PROJECTS.md for how to add or correct an entry via a pull request.