I am having trouble defining this research article exactly how I can reproduce the standard vector quantization algorithm to determine the language of unidentified speech based on a training data set. Here are some basic information:
Abstract information
Language recognition (for example, Japanese, English, German, etc.) Using acoustic functions is an important but difficult problem for current technology speech .... The speech database used in this document contains 20 languages: 16 sentences were pronounced twice by 4 men and 4 women. Each sentence lasts about 8 seconds. The first algorithm is based on the standard Vector Quantization (VQ) method. Each language is characterized by its own codebook the VQ,
.
Recognition Algorithms
The first algorithm is based on the standard vector quantization (VQ) method. Each language kis characterized by its own codebook the VQ,
. At the recognition stage, the input speech is quantized
and the accumulated quantization distortion d_k is calculated. A language that is recognized as minimal distortion. When calculating VQ distortion, several LPC spectral distortions are applied ... in this case, WLR is the weighted smallest ratio - distance: http://tinyurl.com/yc52gcl .
Standard VQ algorithm:
Codebook alt text http://tinyurl.com/y8csx6e , since each language is generated using training sentences. The accumulated distance for the input vector in the sentence
is defined as: alt text http://tinyurl.com/ybynjc2
The distance dcan be any distance that matches the acoustic characteristics, and it should be the same as for codebook generation. Each language is characterized by its codebook the VQ,
.
: ? 50 . MATLAB WLR . , WLR . , VQ 16 ( ) . - , .
!