期刊名称：Signal & Image Processing : An International Journal (SIPIJ)
印刷版ISSN：2229-3922
电子版ISSN：0976-710X
出版年度：2012
卷号：3
期号：1
页码：1
出版社：Academy & Industry Research Collaboration Center (AIRCC)
摘要：This paper presents a technique to synthesize speech from SEMG signals using a frame-byframebasis. SEMG signals are firstly enframed and classified into a number of phonetic classes by aneural network, then the produced sequences of phonetic indices are translated to acoustic signals byconcatenating their corresponding pre-recored speech segments. A significant advantage of the proposedsynthesis based approach compared with previous recognition based approach is that, human is intelligentenough to recognition the synthesized speech although there is errors in it. Experimental evaluations basedon the synthesis of eight words show that on average over 73.4% of the words can be synthesized correctlyand the neural network can classify the SEMG frames of seven phonemes at a rate of 81.9%. The accuracycan be increased to 88.6% by using a glitch removal technique to smooth the produced sequence ofphonetic indices. The results show that the phoneme-frame based speech synthesis technique can beapplied to SEMG-based non-acoustic communication.
