| 摘要 |
A speech detection system for detecting the presence of speech in an audio input signal 1 derived from a predetermined environment such as an HF communication system. Algorithms are used at 4 to extract salient features from a digital signal obtained by analogue to digital conversion at 3 from a received audio signal 1. The algorithm outputs are applied to a neural network 5 which is trained by back propagation to distinguish between speech and other signals. The output of neural network 5 controls a switch or relay 6 to pass an audio signal for the duration a speech is detected. Delay 6 is for ensuring that the initial part of speech is not lost. Filter 10 removes high frequency components introduced by DAC 9. A post processing stage may follow neural network 5 for effective control of switch 6. The algorithm processor 4 may receive a digital audio signal directly as well as via an FFT stage. <IMAGE> |
