Development of speech trainer for the patients wearing maxillary prosthesis

为了使戴上颌骨假体的患者尽快恢复到正常水平，我们试行了一种新的语音训练方法。这一新系统使人们能够使用基于线性预测方程的部分自相关(PARCOR)方法来区分某些元音和塞音。由PARCOR系数表示的声学参数可以识别从前辅音到下一个元音发生转变的部分。该过程包括以下步骤：(1)首先通过单音节模式匹配来区分每个元音，以将计数减少到大小的1/5；(2)选择和记录与所说的单音节最相似的音素。对5名男性说话人的25个元音、305个辅音和130个塞音进行了分析，结果表明，元音和塞音的平均识别率分别为95.4%和78.5%。因此，这一新的系统表明，它可以作为一个重要的语音分析仪，以帮助培训和发展言语疗法。

In order to rehabilitate those patients wearing maxillary prosthesis back to normal standards as soon as possible, a new method of speech training was developed by way of trial. This new system enables one to distinguish certain vowels and plosives using the partial auto-correlation (PARCOR) method, that is based on a linear prediction equation. Acoustic parameters indicated by the PARCOR coefficient can recognize the part where transition from the frontal consonant to the following vowel has taken place. This procedure involves the following steps; (1) each vowel sound is first discriminated by monosyllable pattern matching to reduce the count to 1/5 the size, (2) the phoneme having maximum resemblance to the spoken monosyllable is selected and recorded. The analyzed data comprises 25 vowels and 305 consonants for discriminating vowel, and 130 plosive sounds for recognition of frontal consonants from 5 male speakers.Results showed the recognition rates of the vowels and plosives to be 95.4% and 78.5% respectively on the average. Thus, this new system showed that it could be as an important speech analyzer to aid in training and developing speech therapy.