An investigation for the development of the miniature speech analysis device for linguistic fieldwork

我们研究了适合田野调查的语音分析软件的开发，并为研究收集和编辑了各种语言的语音数据。我们构建了一个通过与数据库中存储的模型语音进行比对来纠正举报人语音错误的系统，并对其进行了声学分析和语音识别技术处理。我们还研究了各种个体数据的归一化方法。一般来说，共振峰和基音的绝对值受声道和声源特征的影响很大，如声道的长度和声带的厚度。考虑到我们已经检查了更多不同数据的归一化，例如共振峰和基音信息，以便通过本方法进行更好的比较。另一方面，我们研究了一种快速的数据处理方法，包括声音归档和声音标注，以便在现场录音后立即建立语音数据库。我们尝试在声学分析的基础上，在图表和横断面上重建声道图像。特别是，我们尝试对重建的声道图像进行图形化处理和呈现，其中包含基于声学分析的元音的舌位和唇形。我们还加强了韵律信息的处理技术，如基音、基频(FO)。等，并试图将它们作为分散的数据进行处理和呈现。此外，我们还尝试使语音分析设备能够仅捕获真实的音调，甚至跟踪音调的突然变化。最后，我们在语音描述的基础上，从音系的角度对所收集的语音数据进行了进一步的考虑。

We have researched on development of speech analysis software, which is appropriate for fieldwork, and collected and edited the speech data of various languages for the research. We have constructed a system which corrects the errors of informant speech by comparing with the model pronunciation stored in database, having analyzed acoustically and processed it with speech recognition techniques. We have also studied on the method of normalization of various individual data. In general, the absolute values of formants and pitches are extremely influenced by the features of vocal tract and sound source, such as the length of vocal tract and the thickness of vocal cords. Considering that we have examined the normalization of more diverse data such as formant and pitch information for better comparison by means of the present method. We have on the other hand, made a research for a fast data processing method, which includes sound filing and sound labeling in order to make a speech database instantly after recording in the field. We have attempted to roconstruct the image of vocal tract on charts and cross sections based on the acoustic analysis of the informants' speech. Particularly, we have attempted to process graphically and present the images of reconstructed vocal tract, which contain tongue positions and lip forms based on the acoustically analyzed vowel sounds. We have also reinforced the processing techniques of prosodic information such as pitch, fundamental frequency (FO). etc, and attempted to process and present them as scattered data. In addition, we have attempted to enable the speech analysis device to capture only the real pitch and even follow the abrupt pitch changes. Finally, we have made an additional consideration on the collected sound data from the phonological point of view, based on the phonetic description of them.