CAJ | 학술논문

依据异类文种之间、同类文种不同语音之间存在音素数据关联的特性，提出多文种语音数据融合编码方法。将不同文种存在的相同音素数据段块按段块模板截取语音样本序列，小波变换，提取特征矢量，生成共享模板集；任意字音或语句音串均按共享模板集提供的元素进行编码与解码；以模板音素串构成的语音记录库按（音节、音素）索引。实验结果表明，单字语音数据压缩比、语音数据存储量、语音还原分段信噪比、主观评价得分等参数均明显优于已有方法，语音还原质量良好。
의거이류문충지간、동류문충불동어음지간존재음소수거관련적특성，제출다문충어음수거융합편마방법。장불동문충존재적상동음소수거단괴안단괴모판절취어음양본서렬，소파변환，제취특정시량，생성공향모판집；임의자음혹어구음천균안공향모판집제공적원소진행편마여해마；이모판음소천구성적어음기록고안（음절、음소）색인。실험결과표명，단자어음수거압축비、어음수거존저량、어음환원분단신조비、주관평개득분등삼수균명현우우이유방법，어음환원질량량호。
Multilingual speech data fusion coding method is proposed according to the phoneme data correlation properties, which exist among heterogeneous languages and different words in the same language. Voice sample sequences of the same phoneme data segment in different languages are intercepted according to the segment templates, wavelet transform is done to those sequences, then feature vectors is extracted to generate shared template sets. Speech data of any word or sentence are coded or decoded according to the template sets. The speech record database made up of template phoneme sets is indexed according to (syllable, phoneme)structure. The single word compression ratio, speech data size, segmental signal-to-noise ratio(SNRS)and score of subjective evaluation(MOS)are significantly better than existing methods, also the voice restoration is of good quality.