论文部分内容阅读
本文以帮助聋儿言语康复为出发点,从聋儿音频发音数据中获得了聋儿易错发音文本以及聋儿易混淆发音文本对。设计了一个数据驱动的3D说话人头发音系统,该系统以EMA AG500设备采集的发音动作为驱动数据,逼真模拟了汉语的发音,从而可使聋儿观察到说话人嘴唇及舌头的运动情况,辅助聋儿发音训练,纠正易错发音。最后对系统的性能进行了人工评测,结果表明:3D说话人头发音系统可以有效地模拟说话人发音时口腔内外器官的发音动作。此外,本文还用基于音素的CM协同发音模型合成的方法,合成了聋儿易错发音文本的发音动动作,并用RMS度量了合成发音动作与真实发音动作的误差,得到了均值为1.25mm的RMS误差值。
Based on the speech rehabilitation of deaf children, this paper obtains the easy-to-mistake pronunciation of deaf children and the confusing phonetic text of deaf children from audio-frequency data of deaf children. A data-driven 3D speaker’s hair sound system is designed. The system uses the phonetic movements collected by the EMA AG500 as driving data to simulate the pronunciation of Chinese so that deaf children can observe the movement of the lips and tongue of the speaker, Assist the deaf children pronunciation training, correct the pronunciation of error-prone. Finally, the performance of the system was evaluated manually. The results show that the 3D speaker’s hair sound system can effectively simulate the pronunciation of the internal and external organs when the speaker sounds. In addition, based on phoneme-based CM synergistic pronunciation model synthesis, the thesis also synthesizes the pronunciation action of deaf children’s erroneous pronunciation texts and measures the error between the synthesized pronunciation action and the real pronunciation action with RMS, and the average value is 1.25mm RMS error value.