| Summary: | Eavesdropping technology has long been of particular interest to security organizations and governments. The improvement of this technology is important for the enhancement of surveillance especially in the face of rising global threats. Its use in commercial applications has also been increasingly explored, in conjunction with the development of automatic speech recognition applications. In this final year project, the effectiveness of beamforming algorithms is explored using automatic speech recognition programs.
This report covers a review of beamforming concepts and signal processing techniques that are useful in the application of speech recognition. The effect of different parameters on beamforming is also explored by simulating an algorithm in a virtual room environment. Audio files containing up to two interference channels were created using audio editing software. Using both planar and circular microphone arrays, a setup was created to record audio in a meeting room environment. The recordings were transcribed using current speech recognition software and analysed using a program that calculates the word error rates by comparing the reference and hypothesis scripts.
It is concluded that beamforming algorithms are effective in enhancing the desired signals based on the results obtained. The circular microphone array also produced better performance. Further analysis may be conducted by changing the recording environment and developing a new speech recognition system for far-field signals.
|
|---|