音訊信號的隨機性程度測試

Abstract

摘要

本次實驗的目的是要分析音訊信號中不同片段的隨機性程度，並判斷哪些片段屬於隨機信號或是不屬於隨機信號，最後討論隨機信號跟語音中的清音(unvoiced sound)與濁音(voiced sound)特徵的關係，也會討論取出片段的長度對於隨機信號判斷的關係。

本次實驗是根據課本第116頁的方法，先標記音訊信號中所有的轉折點，再分片段統計片段中轉折點的個數，最後利用音訊信號中局部片段的轉折點個數佔此片段的比例來判斷此片段的信號是否為隨機信號。 本次實驗使用的測試信號為兩段錄音，音訊內容分別為"safety"與"吳宜峰"，錄音的工具是手機(OPPO R11s)，音訊採樣頻率為44100 Hz。

觀察結果後可以發現特徵跟理論大致有符合。聲音如果是unvoiced sound與plosive sound的話，轉折點的個數總和會較大，若聲音是voiced sound，轉折點的個數總和會較小。

Abstract

The purpose of this experiment is to analyze the degree of randomness of different segments in the audio signal, and determine which segments are random signals or not, and finally discuss the relationship between random signals and the unvoiced and voiced features in speech, and will also discuss the relationship between the length of the extracted segment and the judgment of random signals.

This experiment is based on the method on page 116 of the textbook. First, mark all the turning points in the audio signal, and then count the number of turning points in the segment by segment, and finally use the proportion of the number of turning points in the partial segment of the audio signal to the segment to determine whether the signal of this segment is a random signal. The test signal used in this experiment is two recordings, the audio content is "safety" and "Wu Yi-feng", the recording tool is a mobile phone (OPPO R11s), and the audio sampling frequency is 44100 Hz.

After observing the results, it can be found that the characteristics are roughly consistent with the theory. If the sound is unvoiced sound and plosive sound, the total number of turning points will be larger. If the sound is voiced sound, the total number of turning points will be smaller.