Multichannel Crosstalk Resistant ANC

Get Complete Project Material File(s) Now! »

Combined MCRANC with delay and sum beamforming

In this section, a combined algorithm of MCRANC and Delay And Sum (DAS) beamforming is presented for speech enhancement. It first employs MCRANC for every channel of the noisy speech to get enhanced speeches with higher SNR and less correlated residual noises. Then the enhanced speeches are input to a DAS beamformer to get further enhancement. 3.3.1 Delay and sum beamforming The DAS algorithm is the most basic algorithm for beamforming. It aligns the array signals and then sums the aligned signals to get the output. Its structure is indicated in figure 3.3.1.

Capability of DAS beamforming

In the ideal condition that the noises are completely uncorrelated and the time alignments are precise, it can be proved that the SNR improvement provided by DAS beamforming is SNRimproved= ) 10log ( 10 N (3.3.10) where N is the number of the microphones in the array. Since 10log (2 ) 3 10log ( ) 10 N ≈ + 10 N the SNR will increase about 3 dB as the number of the microphones doubles. However, the SNR improvement will greatly decrease as the noise correlation increases. In fact, DAS will not provide any SNR improvement if the noise correlation reaches its maximum value 1. For a small microphone array, the noises in the microphone signals are more highly correlated and fewer microphones can be employed. As a result, DAS can provide only very limited SNR improvement to small microphone array. To get better SNR improvement, it should be used with other algorithms.

Combined MCRANC with DAS beamforming

As shown in figure 3.3.2, the combination scheme of MCRANC with DAS consists of N subsystems of MCRANC and a DAS beamformer, where N is the number of microphones employed in the array. Every dot-lined frame in figure 3.3.2 contains an N-input and one-output MCRANC subsystem. The output of any MCRANC subsystem is actually a primarily enhanced speech signal. These enhanced speech signals are input to DAS beamformer to get further enhancement. Figure 3.3.3 indicates the i-th MCRANC subsystem, in which the i-th channel of signal is used as the main signal and the other N-1 channels of signals are used as referential signals in a MCRANC.

MCRANC subsystem

The subsystem as shown in figure 3.3.3 is a MCRANC as introduced in chapter 2. However, the index notations for the subsystem are somewhat different from those in chapter 2. Unlike chapter 2, there are only N channel noisy signals in this section and every channel of the signal is used in turn as a main channel signal while the other N-1 channels are treated as referential signals. Because these MCRANC subsystems will also be referred to in the next sections and the following chapters, they are described again as follows. Suppose speech s(k) and noise n(k) are generated by independent sources. They arrive at microphone M i through multi-paths and are acquired by M i as s (k) i and n (k) i respectively. The impulse responses of the intermediate media between the speech and noise sources and the microphone M i are h (k) si and h (k) ni respectively.

Adaptive module controller

From equations (3.3.18) to (3.3.23) we know that the impulse response of filter Ai is * wi . With filter Ai we can cancel the noise n (k) i carried with the signal x (k) s (k) n (k) i = i + i . If the transfer function or impulse response of the noise is unchanged, which implies the whole environment remains unchanged including the positions of the noise sources, the space and even the air temperature and pressure, the optimal weight * wi would remain unchanged. But unfortunately the transfer function of the noise keeps changing from time to time with the changes of environment such as the opening of a door or the closing of a window. To adapt the system to the dynamical changes of the environment, the weights of filter Ai must be adapted from time to time during NSP time sections to compensate for any change in the noise environment.

Contents :

• Chapter 1 Introduction
  • 1.1 Speech enhancement
  • 1.2 Research history of speech enhancement
  • 1.3 Introduction to speech enhancement algorithms
    • 1.3.1 Single-channel speech enhancement algorithms
    • 1.3.2 Microphone array speech enhancement algorithms
  • 1.4 Evaluation of the enhanced speech
  • 1.5 Strategies and relevant work
  • 1.6 Thesis contributions
  • 1.7 Structure of thesis
• Chapter 2 Multichannel Crosstalk Resistant ANC
  • 2.1 Introduction
  • 2.2 Adaptive noise cancellation
    • 2.2.1 Two-channel adaptive noise cancellation
    • 2.2.2 Multichannel adaptive noise cancellation
  • 2.3 Two-channel crosstalk resistant ANC
    • 2.3.1 Algorithm and its principal
    • 2.3.2 Adaptation algorithm
  • 2.4 Multichannel crosstalk resistant ANC
    • 2.4.1 MCRANC algorithm
    • 2.4.2 Computational complexity
  • 2.5 Experimental results
    • 2.5.1 Simulation experiment
    • 2.5.2 Experiments in real environments

• Chapter 3 Combined Algorithms With MCRANC
  • 3.1 Introduction
  • 3.2 Combined MCRANC with improved spectral subtraction
    • 3.2.1 Description
    • 3.2.2 Improved spectral subtraction
    • 3.2.3 Experimental results
    • 3.2.4 Conclusions
  • 3.3 Combined MCRANC with delay and sum beamforming
    • 3.3.1 Delay and sum beamforming
    • 3.3.2 Combined MCRANC with DAS beamforming
    • 3.3.3 Experimental results
    • 3.3.4 Conclusions
  • 3.4 Combined MCRANC with Weiner post-filtering
    • 3.4.1 Weiner post-filtering
    • 3.4.2 Combined MCRANC with Weiner post-filtering
    • 3.4.3 Experimental results
    • 3.4.4 Conclusions
  • 3.5 Summary
• Chapter 4 Improved MCRANC Methods
  • 4.1 MCRANC with multichannel distorted signal filtering
    • 4.1.1 Description of the method
    • 4.1.2 Combined with DAS beamforming
    • 4.1.3 Comments
    • 4.1.4 Experimental results
    • 4.1.5 Conclusions
  • 4.2 MCRANC using multiple sampling rates
  • 4.3 Fixed beamforming partial-channel MCRANC
    • 4.3.1 Fixed beamforming MCRANC
    • 4.3.2 Delay and weighted sum beamforming
    • 4.3.3 Partial-channel MCRANC
    • 4.3.4 Fixed beamforming partial-channel MCRANC
    • 4.3.5 Experimental results
    • 4.3.6 Conclusions
  • 4.4 Subband MCRANC
• Chapter 5 Improved MGSC Algorithms
• Chapter 6 Hybrid Algorithms
• Chapter 7 Conclusions And Future Work
  • References

GET THE COMPLETE PROJECT
Speech Enhancement Using A Small Microphone Array