以OMAP5912實現語者方位判定於輪式機器人平台

Abstract

本論文以聲源方位判定演算法(DOA, Direction of arrival)判斷語者方位，其中結合了聲源啟動偵測系統(VAD, Voice Activity Detection) 以及聲源方位判定系統(GCC, Generalized Cross-Correlation)應用於機器人聽覺系統，由這兩個聲音處理系統整合成為DOA演算法。本理論預期要以OMAP5912作為機器人的核心系統，在其上架設雙聲道麥克風，偵測環境中的聲源以及噪音源。VAD系統主要針對環境偵測是否存在語音，並且在偵測語音後，紀錄語音頻帶，提出新的係數調整演算法，傳送到下級GCC系統計算出聲源角度，可適應即時環境噪音之改變。因為OMAP5912平台包含有DSP5510系統核心以及ARM系統核心，所以論文初期以PC驗證演算法，進而以DSP平台DSK5510做驗證，後期預計整合進OMAP5912中並實際與輪式機器人整合，為系統增加智慧型功能，可應用於多樣環境追蹤人聲以及環境監測等功能。

This thesis proposes a DOA algorithm (Direction of arrival) used in source direction detection system-GCC (Generalized Cross-Correlation) and voice activity detection (VAD) system for applications in robot where speech reception is contaminated by various noise and sound sources. Two microphones connected with TI OMAP5912 were made for the implementation platform. To cope with the changes of noise and environmental characteristics, the adaptive VAD system with on-line noise channel calibration is designed and implemented. The VAD result is used to ensure the correct timing for on-line calibration and choose the voice’s frequency without consulting the noise’s. The algorithms are implemented on OMAP platform with DSP and ARM dual core. To verify the effectiveness of the proposed system, the speech signal and angel of voice are combined with robot system that demo the movement of robot. The experimental results show that the algorithms are able to detect the direction of source and reduce the error of angel.