這篇論文研究的是如何將繪製幾何圖形的口語表達轉換為繪圖軟體GeoGebra的指令。透過語音辨識，將聲音轉換為文字，從文字中分析並組合出符合的GeoGebra繪圖指令，並使用自動化程式操作GeoGebra，使其自動繪製幾何圖形。

自動化繪圖系統的流程主要分成三個階段：語音辨識、文字轉指令、執行GeoGebra。語音辨識引用python的相關套件，將聲音轉成文字。文字轉指令是這份論文最主要的部分，轉換方式有兩個流程。第一個流程是先將文字分類出所代表的繪圖指令，依據選定的繪圖指令設定關鍵字，記錄文字裡出現關鍵字的次數和位置，將文字以向量表示，藉此觀察文字在多維空間的分佈情形。定義兩個文字之間的距離函數，判斷文字在多維空間中的距離遠近，並使用K-近鄰演算法進行分類。第二個流程是透過各個繪圖指令的語意分析程式，將所需要的參數組合成完整的繪圖指令。第三個流程就是透過自動化程式執行GeoGebra，自動產生圖形。

這三個階段都需要考慮判斷的準確率，為了模擬語音辨識錯誤的情形，在提供給K-近鄰演算法的資料中，也收集錯字、缺漏字等不完整的句子。在文字轉指令的過程中，我們有84%的準確率。然而在錯誤的情況中，大多是因為口語文字接近其他繪圖指令類別，導致K-近鄰演算法分類錯誤。

未來可語音控制的自動化繪圖系統提供使用者一個新的繪圖方式，結合語音辨識、機器學習與自動化控制，讓繪圖軟體的操作過程更加創新。

This paper focuses on how to convert the spoken expressions of drawing geometric figures into instructions for the drawing software GeoGebra. By speech recognition, the sound is converted into text, and the corresponding GeoGebra drawing instructions are analyzed and combined from the text. GeoGebra is operated by an automated program to automatically draw geometric figures.

First, we introduce the progress of our system. There are three stages: speech recognition, text-to-command, and GeoGebra execution. The second stage is the most important stage in this system. There are two processes in this part. The first process is consist of (i) classifying the text into the drawing instructions represented, (ii) setting keywords according to the selected drawing instructions, (iii) recording the number and position of the keywords in the text, and (iv) representing the text as a vector to observe the distribution of the text in the multi-dimensional space situation. We will introduce GeoGebra and some packages of Python, and k-nearest neighbors algorithm. We define the vector space and distance that we need, then use them for our model.

In these three stages, the accuracy of judgment needs to be considered. In the process we have 84% accuracy. However, in the wrong cases, mostly because the spoken words are close to other drawing instruction classes, the K-Nearest Neighbors algorithm classifies them incorrectly.

In the future, the automatic drawing system will provide users with a new drawing method, combining voice recognition, machine learning and automatic control to make the operation process of the drawing software more innovative.

[1] SAADATFAR, Hamid, et al. A new K-nearest neighbors classifier for big data
based on efficient data pruning. Mathematics, 2020, 8.2: 286.
[2] ZHANG, Zhongheng. Introduction to machine learning: k-nearest neighbors.
Annals of translational medicine, 2016, 4.11.
[3] GUO, Gongde, et al. KNN model-based approach in classification. In: OTM
Confederated International Conferences” On the Move to Meaningful Internet
Systems”. Springer, Berlin, Heidelberg, 2003. p. 986-996.
[4] ZULNAIDI, Hutkemri; ZAKARIA, Effandi. The effect of using GeoGebra on
conceptual and procedural knowledge of high school mathematics students.
Asian Social Science, 2012, 8.11: 102.
[5] SANGWIN, Chris. A brief review of GeoGebra: dynamic mathematics. MSor
Connections, 2007, 7.2: 36.
[6] ARBAIN, Nazihatulhasanah; SHUKOR, Nurbiha A. The effects of GeoGebra
on students achievement. Procedia-Social and Behavioral Sciences, 2015, 172:
208-214.
[7] CLANCEY, William J. Classification problem solving. Stanford, CA: Stanford
University, 1984.
[8] HAGHIGHI, Sepand, et al. PyCM: Multiclass confusion matrix library in
Python. Journal of Open Source Software, 2018, 3.25: 729.
[9] HONG, Chong Sun; OH, Tae Gyu. TPR-TNR plot for confusion matrix. Com-
munications for Statistical Applications and Methods, 2021, 28.2: 161-169.
[10] GUNDECHA, Unmesh. Learning Selenium testing tools with Python. Packt
Publishing Ltd, 2014.