With the aid of speech enhancement and noise reduction techniques the clarity and quality of speech communication and recognition systems can be improved. Adaptive Noise Reduction (ANR) is an evolution from single-channel noise suppression algorithms. Adding a second microphone that is able to sample the noise of the acoustic environment, then that signal can be subtracted from the original microphone. The solution now becomes an adaptive filtering solution, which estimates the transfer function between the two microphones. ANR often is a better solution because the noise spectrum can change quickly over time, but the transfer function typically changes much more slowly.
Dual channel ANR is an system identification adaptive filtering problem, very similar to line and acoustic echo cancellation. As shown in Figure below , one input source, s1(n), contains a desired signal and a linear transformed copy of an the noise source. The second input source, s2(n), contains a reference of the noise source signal. An adaptive filter can be applied to the s2(n) and subtracted from s1(n) to remove noise source from s1(n) to generate e(n). The noise source in signals s2(n) is correlated and linear transformed version of the noise source in s1(n), so optimal filter can be found which will minimize the output power of e(n).
VOCAL’s ANR is available as a standalone module or as part of a comprehensive VoIP System customizable to most platforms and environments. The ANR is available for several different platforms, including true DSPs such as Texas Instrument’s 64x, 62x, 55x, and 54x families as well as ADI’s Blackfin, SHARC and 218x families. General-purpose CISC/RISC architectures such as PowerPC, PowerQUICC, x86, ARM, and MIPS are also supported.