The Study Of Baseband Extraction In The Noisy Environments

With more and more people demand of multimedia information, audio occupy an increasingly important position in the multimedia resources, including humming, including audio retrieval technology become a hot research. Humming, the accuracy and robustness of the pitch frequency due to noise, humming level factors will be restricted. The research papers address these issues, the main work includes:1. Traditional baseband algorithm and comparative experiments. Major is research autocorrelation function method, the inverted spectrum method algorithms, comparative experiments on different humming and ambient noise. Humming vocal range change and environmental noise for a great influence on the accuracy of the algorithms extract baseband.2. Research baseband extraction algorithm based on linear prediction residual cepstrum and two aspects of the on Voicing judgment and high frequency set zero by the line improvements. Experiments verified that, the combined linear prediction cepstral analysis method to solve the "robustness" the shortcomings of the traditional cepstrum, accuracy and robustness of a good tradeoff.For Voicing decision algorithm based on linear prediction residual cepstrum around the zero-rate and short-term energy to do the research, the paper focuses on the combined threshold judgment method and the high and low frequency judgment, vowel Judgments Act. The melody contour usually described by the baseband is the most impressive people, the melody contour method can effectively solve the rhythm changes and humming by the range of the specific changes in the scope of the difficulties caused by the fundamental frequency of the vowel curve and music identical melody contour, after experimental verification, the accuracy and robustness of the algorithm extract baseband has a better improvement.3. Study a combination of speech enhancement in pitch extractionIn order to reduce the impact of noise on the baseband extracted, using a speech enhancement of audio preprocessing, and then extracted baseband. Traditional speech enhancement method is the use of the low-pass filter to filter out high frequency noise, but this will filter out the noise while fuzzy audio, which affects the fundamental frequency extraction accuracy. Papers from baseband error rate and standard deviation of the two fundamental frequency extraction algorithm and noise reduction pretreatment method with select effective baseband. Different combinations by line comparison experiments to verify the excellent noise reduction preprocessing to select matches the fundamental frequency extraction algorithm.