Abstract:
Arrays of simple, unmatched microphones have been applied to two speech processing problems. The first application is noise reduction via robust constrained beamforming. Significant noise reduction has been observed (10 dB) in typical rooms at a source-to-array distance of 1 m. The noise reduction achievable is proportional to the direct-to-reverberant energy ratio of the interfering sound energy (from a single location in the room). The robust adaptive processing avoids the cancellation of the desired signal in all cases examined. The second application is speech coding. A standard code-excited-linear-prediction (CELP) codec is used in conjunction with the microphone array and noisy speech inputs. The quality of the desired speech coded in the presence of acoustic interference is enhanced by the array processing. Signal-to-noise ratios, intelligibility averaged gains, other spectral measures, and spectral envelope measures have been compared to assess the relative quality of the spatially filtered and coded noisy speech. [Work supported by NSF Grant No. IRI-9405286.]