A new connected word recognition algorithm based on HMM/LVQ segmentation and LVQ classification

Author(s):  
P. Ramesh ◽  
S. Katagiri ◽  
C.-H. Lee
2013 ◽  
Vol 5 (2) ◽  
pp. 101-104
Author(s):  
Tomyslav Sledevič ◽  
Liudas Stašionis

The paper describes the FPGA-based implementation of Lithuanian isolated word recognition algorithm. FPGA is selected for parallel process implementation using VHDL to ensure fast signal processing at low rate clock signal. Cepstrum analysis was applied to features extraction in voice. The dynamic time warping algorithm was used to compare the vectors of cepstrum coefficients. A library of 100 words features was created and stored in the internal FPGA BRAM memory. Experimental testing with speaker dependent records demonstrated the recognition rate of 94%. The recognition rate of 58% was achieved for speaker-independent records. Calculation of cepstrum coefficients lasted for 8.52 ms at 50 MHz clock, while 100 DTWs took 66.56 ms at 25 MHz clock. Article in Lithuanian. Santrauka Pateikiamas lietuvių kalbos pavienių žodžių atpažinimo algoritmo įgyvendinimas lauku programuojama logine matrica (LPLM). LPLM įrenginys pasirinktas dėl lygiagrečiai veikiančių procesų įgyvendinimo galimybės taikant VHDL kalbą. Tai užtikrina spartų signalų apdorojimą esant taktiniam dažniui iki 50 MHz. Kalbos požymiams išskirti taikoma kepstrinė šnekos analizė. Požymiams palyginti taikomas dinaminis laiko skalės kraipymo (DSLK) metodas. Sudaryta 100 žodžių požymių biblioteka, kuri saugoma vidinėje LPLM BRAM atmintyje. Pasiektas 94 % atpažinimo tikslumas priklausomai nuo kalbėtojo ir 58 % – nepriklausomai nuo kalbėtojo. Kepstro koeficientų skaičiavimas vienam žodžiui trunka 8,52 ms, esant 50 MHz taktiniam dažniui, ir šimtui DLSK – 66,56 ms, esant 25 MHz taktiniam dažniui.


1996 ◽  
Vol 5 (1) ◽  
pp. 67-78 ◽  
Author(s):  
Kenyatta O. Rivers ◽  
Linda J. Lombardino ◽  
Cynthia K. Thompson

The effects of training in letter-sound correspondences and phonemic decoding (segmenting and blending skills) on three kindergartners' word recognition abilities were examined using a single-subject multiple-baseline design across behaviors and subjects. Whereas CVC pseudowords were trained, generalization to untrained CVC pseudowords, untrained CVC real words, untrained CV and VC pseudowords, and untrained CV and VC real words were assessed. Generalization occurred to all of the untrained constructions for two of the three subjects. The third subject did not show the same degree of generalization to VC pseudowords and real words; however, after three training sessions, this subject read all VC constructions with 100% accuracy. Findings are consistent with group training studies that have shown the benefits of decoding training on word recognition and spelling skills and with studies that have demonstrated the effects of generalization to less complex structures when more complex structures are trained.


Author(s):  
Manuel Perea ◽  
Victoria Panadero

The vast majority of neural and computational models of visual-word recognition assume that lexical access is achieved via the activation of abstract letter identities. Thus, a word’s overall shape should play no role in this process. In the present lexical decision experiment, we compared word-like pseudowords like viotín (same shape as its base word: violín) vs. viocín (different shape) in mature (college-aged skilled readers), immature (normally reading children), and immature/impaired (young readers with developmental dyslexia) word-recognition systems. Results revealed similar response times (and error rates) to consistent-shape and inconsistent-shape pseudowords for both adult skilled readers and normally reading children – this is consistent with current models of visual-word recognition. In contrast, young readers with developmental dyslexia made significantly more errors to viotín-like pseudowords than to viocín-like pseudowords. Thus, unlike normally reading children, young readers with developmental dyslexia are sensitive to a word’s visual cues, presumably because of poor letter representations.


Sign in / Sign up

Export Citation Format

Share Document