Howling Detection Method In Television And Telephone Conference System

As the development of Internet communication technology and economic conditions, the television and telephone conference with its own advantages has been more and more widely used, and it has become a kind of office operation and communication mode that is generally accepted by people. For the strong interactive application of television and telephone conference, the guarantee of multimedia transmission quality is the key to the operation of the system. However, because of the influence of specific environmental factors and the bidirectional communication mode in television and telephone conference, it is extremely easy to form a closed loop in the system, when the closed loop circuit meet howling occurred conditions of amplitude balance and phase equilibrium, the system output signal will form a self-excited oscillation, formation acoustic feedback. The howling will result in extremely bad environment, even destroy the sound reinforcement components makes it unable to work properly, therefore, we should pay attention to the howling dangers for the television and telephone conference system, in order to have a good sound amplification effect we need to do howling detection and suppression work in the acoustic scene to ensure the audio quality and the basic meeting environment.This subject comes to project sound detection system, the system do the real-time monitoring and collection of audio information in the conference hall of television and telephone conference, when the conference system in squeaking, the howling signal after the growing process of short time reach saturation state, in this state, the signal power output is the biggest, time domain waveform will appear serious cut the top of the phenomenon and the frequency-domain will have lots of harmonics appear, at this time if the howling detection, test results will appear in large deviation, and the reaction time will slowly. Consider did not reach the saturated state of the howling signal, this time in the growth state and time domain waveform whose amplitude value continues to increase, the frequency domain performance for single frequency and power increasing similar to the sine wave signals. At this time if the howling detection will achieve the desired effect, but the signal duration is very short, which proposed requirements for howling detection.This paper has analyzed characteristics of the digital speech signal mixed with howling which picked from sound detection system, from the frequency domain analysis, wavelet analysis two angles were howling detection and prediction of howling occurrence frequency and detection to the howling formed the initial time, efficiency and response speed and verification method of the executive. The speech signal is non-stationary signal, and the traditional Fourier analysis method is not directly applicable to the signal, but within a short range of the speech signal can be used as a quasi-steady state process, with short smooth characteristics.From the perspective of the frequency domain detected the howling signals, it must first be pretreated, and will be framed for the short-time stationary characteristic of the speech signals. Through a fixed length of time window in a fixed frame shift length of the original signal is segmented, frame length and frame shift decision signal processing time resolution and frequency resolution, the choice of different windows have different bandwidths and spectral leakage will affect the result of the voice analysis. Based on the short time frame use the short time Fourier transform to analyze the frequency domain, each frame of the speech signal can be considered to be intercepted from different stationary signal waveform out, short-term spectrum of each frame of speech is the individual smooth waveform spectrum approximation. According to the obtained short-time signal spectrum, this paper from the energy, phase and time three aspects setting control conditions to detect howling, to squeal frequency energy accounts for the frame energy ratio of the larger as the basis to determine each frame peak to average power ratio maximum points as pending squeal frequency; acoustic feedback condition of phase equilibrium, the adjacent frame between squeal frequency phase difference is constant, the phase spectrum frame between the second order variation will fall in the near to zero; in short time duration howling formation process, if the continuous N frames can meet the energy and phase control conditions is judged in this case in the howling.By the short-time Fourier transformation process, it is known that it's a single resolution analysis method, the low frequency and high frequency part of the signal is decomposed with fixed time-frequency resolution and frequency resolution. The speech signal is non-stationary signal, a normal voice signal energy mainly concentrated in the low frequency components, and the howling part usually is a narrow band single frequency and high frequency signal. Wavelet analysis has good time-frequency local analysis ability, the low frequency part of the signal with low time resolution and high spectral resolution and low frequency resolution and high time resolution in high frequency part. From the time domain waveforms of each frequency band signal which obtained by wavelet packet decomposition of signal to find the howling signal.