Font Size: a A A

Research On 8~32kb/s Wideband Embedded Variable Bit Rates Speech Codec

Posted on:2008-01-24Degree:MasterType:Thesis
Country:ChinaCandidate:Z X LiuFull Text:PDF
GTID:2178360215994809Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
The embedded variable bit rate speech coding has become one of the most interesting international issues with the rapid development of IP transmission and the great application of IP telephone. At present, the International Telecommunications Union Standardization Sector is studying an embedded variable bit rates speech coding standard named as G.VBR. Under this background, a five layers wideband embedded variable bit rate speech codec is proposed based on ACELP (Algebraic code excited linear prediction) and TCX (transform coded excitation) techniques. The codec has been submitted to ITU-T as one of four international candidates.According to the terms of reference of G.VBR speech coding standard from ITU-T, a five embedded variable bit rates speech codec at 8~32kb/s only based on Algebraic code excited linear prediction technique is proposed firstly. In this codec, the core layer adopts the basic ACELP coding model, and the enhancement layers (from 12kb/s to 32kb/s) are obtained by increasing the number of pulse. We adopt muti-stage target vectors in order to obtain the embedded relationship among all the layers, meanwhile, separate adaptive codebooks are introduced and the memories of the filters are updated separately to match the parameters of the different layers better.Secondly, in order to improve higher layers'speech quanlity, Instead of the embedded ACELP coding model, the embedded TCX coding technique is used to operate the enhancement layer three and enhancement layer four, in order to obtain the complete embedded structure in the all five layers, the difference between the pre-processed speech and the local synthesis speech of 16kb/s is regarded as the target signal of enhancement layer three, and the difference between the unquantized and the quantized target of enhancement layer three is regarded as the target signal of enhancement layer four. At the same time, the perceptual weighting filter is improved in the embedded TCX coding. Replacing the algorithm of choosing three among four pulses, Focused-Search algorithm is adopted to search the Algebraic codebook.The proposed codec with the bit rates from 8kb/s to 32kb/s by using ACELP and TCX techniques has matched the structure of ITU-T. The test results indicate the codec has good quality and low delay.
Keywords/Search Tags:Speech Coding, Embedded Speech Coding, ACELP, TCX
PDF Full Text Request
Related items