Research On Phonetic Similarity Evaluation Algorithm

With the improvement of the people’s material and spiritual culture more and moreabundant, many singing contest program ratings link is an integral part. But Most ofthe scoring method is the authority of the judges in the field to decide the basis of theirprofessional skills, as well as inside and outside audience support players’ performanceresults. The results of this rating will inevitably mixed personal subjective factors ofthe judges and the audience, there is a dispute arising due to personal preferences.Existing scoring software may be due to the unreasonable rated parameters and limitedsimilarity matching, made the final score is not very satisfactory. In order to solve theproblem, looking for a quick and effective voice evaluation method is particularlyimportant.Voice similarity evaluation is an automatic evaluation method for the degree ofsimilarity voice of original singers and imitators, Its process is to extract featureextraction and similarity matching from input standard voice and the imitate voice.Make the computer simulation its Subjective impression for similarity to evaluate.To ensure the accuracy of speech recognition voice is the premise for an accurateassessment. Voice pre-emphasis, a windowing function, framing a series ofpre-treatment is the first; the purpose is to reduce the background noise in the voice, toimprove the purity of the voice. Use the method of combining short-time averageenergy and short-time average zero-crossing rate for voice endpoint detection featureextraction, then sequence the characteristic parameters into pattern matching, and cameto the conclusion rated. Characteristic parameter extraction and similarity matching isthe core of the voice similarity evaluation system. In Feature extraction, select thepitch frequency, Mel cestrum coefficient and the intensity of the sound as a voicefeature similarity evaluation parameters; In similarity matching, this paper uses ant colony dynamic programming algorithm to find the optimal path of a match between aspeech feature parameter sequence, and then using the Euclidean distance to measurethe similarity of voice. The littler average match distortion (the difference betweenthe two) the higher scores of imitators. Experimental results show that the system has ahigh accuracy in imitate voice and original voice on the degree of similarity inpronunciation and intonation.