CAJ | 학술논문

为了提高噪声环境下的语音、声调语音以及音乐的识别水平，基于希尔伯特黄变换提出一种新的电子耳蜗语音编码策略，利用经验模态分解和希尔伯特变换提取语音的幅度瞬时幅度和瞬时频率，经滤波、调制等处理算法获取表征语音的精细结构进而合成刺激信号。以Matlab软件为平台对提出的算法和传统的连续间隔采样以及幅频联合编码算法进行仿真，分别处理50组语音测听材料并合成相应的语音信号。结果显示新算法合成的语音信号与原始信号的相关系数高于另外两种算法得到的相关系数，从而表明新算法可以保留更多原始语音的信息。
위료제고조성배경하적어음、성조어음이급음악적식별수평，기우희이백특황변환제출일충신적전자이와어음편마책략，이용경험모태분해화희이백특변환제취어음적폭도순시폭도화순시빈솔，경려파、조제등처리산법획취표정어음적정세결구진이합성자격신호。이Matlab연건위평태대제출적산법화전통적련속간격채양이급폭빈연합편마산법진행방진，분별처리50조어음측은재료병합성상응적어음신호。결과현시신산법합성적어음신호여원시신호적상관계수고우령외량충산법득도적상관계수，종이표명신산법가이보류경다원시어음적신식。
To enhance speech recognition in realistic listening environment, as wel as tonal language and music perception, a new speech coding strategy based on Hilbert Huang transform was presented. Instantaneous frequency and instantaneous amplitude which reflect speech contents, speech rhythms and tones are derived from original speech signal through empirical mode decomposition and the Hilbert transform to synthesize stimulating pulses. The presented new speech coding algorithm, continuous interleaved sampling, and frequency amplitude modulation encoding strategies were simulated by Matlab and synthesized signals of 50 Mandarin speech test materials are correlation analyzed between original signals. Compared to other two strategies, the presented new strategy obtains the highest correlation coefficient between synthesized signal and that of original speech, which indicates it could keep more information of the original speech signal than other two strategies.