There has been a tremendous growth using digital multimedia in the last two decades. Speech and audio contents have always been an integral part of multimedia. The exponential growth in demand for digital audio has created requirements for high fidelity content as the communication network are growing. Consequently, Digital multimedia has lead to several international standards for speech and audio convergence codecs such as AMR-WB+ codec.
Speech codecs are based on vocal tract modeling for redundancy removal using a technique called Linear Predictive Coding. The human vocal tract is modeled as a time varying filter. The parameters of the filter are transmitted or stored. Speech codecs are highly effective for single speaker material with the signals sampled at 8 kHz or 16 kHz. Their round-trip algorithmic delay less than 30 ms to make codecs useful for full-duplex communications purposes. On the other hand, audio codecs achieve compression by exploiting the perceptual redundancies present in the audio signal. This perceptual redundancy is due to the limitations of the human auditory system.
The convergent codec of speech codec and audio codec based on different coding algorithm are hard to converge using unified coding algorithm completely. Recently, in the field of mobile communication, multi mode codec has been standardized for handle to signals of different characteristics between speech and audio.
In this paper, a novel approach is proposed to improve the performance of audio/speech classification using MFCC based on GMM probability model used for the Media Classification System.
By using the Open-loop and use storage buffer using the correlation with the past frame not only reduces the amount of calculation existing but also the signal enhanced accuracy of the classification. In particular, it was found to show improved performance in the music signal such as Castanet and symbols are not good signal classification of common USAC. The performance of the proposed classification algorithm using MFCC feature vector and storage buffer was reduced the error rate about 10% and Computational complexity was also reduced.
The simulation results show that the performance of the proposed signal classifier algorithm has better performance than the conventionally implemented USAC scheme.