Active noise reduction (or cancellation) in sound capture can be performed in several different manners and is dependent on the information that is available. In the acoustic environment there are many sources of noise consisting of different spectrum characteristics, either time-invariant or time-varying. In almost any application related to sound capture and audio communication, the microphone not only picks up the signal of interest, but also picks up other audio sources that lower the quality of the signal of interest for the far-end user.
Single-channel Noise Reduction
The classical approach for a single-channel active noise control application is spectral subtraction. In spectral subtraction, the idea is that the acoustic noise is additive and thus can be subtracted from the disturbed signal. This concept is relatively simple but relies solely upon an accurate estimate of the noise spectrum. Therefore, the main criterion of a single-channel method is obtaining an accurate time-variant model of the system.
In more modern systems, components related to speech are also estimated and enhanced to provide a higher overall SNR. For example, some speech enhancement software routines estimate the pitch frequency of a voiced section of speech and then apply a comb filter to remove spectral components not associated with the pitch frequency and its harmonics.
In acoustic environments in which the sound source and microphone are located some distance apart, reverberation is most likely going to be a problem. Because we do not have access to the original sound source, dereverberation via deconvolution filters a model of the room impulse response from the received signal.
One approach to this acoustic noise reduction problem is to use linear prediction analysis. Based on the knowledge of the human speech production system, only the LP residuals will be affected by reverberation. This information can than be applied to an update of the adaptive filter for deconvolving the room impulse response from the received signal.
Multi-channel Noise Reduction
In multichannel scenarios for active acoustic noise cancellation, usually a second microphone is placed near the noise source and far enough away from the sound source so the signal from the second microphone can be used as the estimate in the spectral subtraction technique described above.
In acoustic beamforming the spatial relationship between the multiple microphones is used for active noise control. If the direction of sound source relative to the microphone array is known then an acoustic beamformer can be designed to pass signals coming from the sound source and filter out signals coming from different directions. This approach is most applicable to a situation in which multiple people are talking but only one is desired to be heard.
In a multichannel system where the direction of arrival information is unreliable, a Dual-Channel estimation technique can be used for acoustic noise cancellation. This is an extension of the single channel estimation method and uses the phase difference information to determine the probability of the presence of speech and to estimate the speech spectrum components.
Additional information is also available for:
- Noise Reduction Using Minimum Mean Square Estimators
- Reduction of Noise in Non-stationary Sources
- Psychoacoustic Noise Suppression
- Model-Based Speech Enhancement
- Musical Noise in Acoustic Noise Reduction
- An Instructional Example
