To recover speech corrupted by functional magnetic resonance imaging (fMRI) acoustic noise using two-channel adaptive speech enhancement techniques. Speech corrupted by noise generated from a 3 T MRI scanner was recorded using diffuse-field microphones and a data acquisition board. Multiband and subband adaptive speech enhancement methods are used to recover the speech signal from the recordings. Normalized least mean squares (NLMS) algorithm was used for updating the filter coefficients in each band. The methods are successful in enhancing the speech quality. They are successful in improving the convergence rate of the adaptive filter. Multiband and subband methods have a similar performance in terms of noise reduction and in the subjective tests. The subband method introduces less speech distortion compared to the multiband method. The subband method requires a lower number of computations per sample. Adaptive speech enhancement techniques are effective in reducing fMRI background noise in the recordings. Based on the analysis, we conclude that subband-based methods are more suited for enhancing speech corrupted by fMRI noise.