應用適應性演算法於語音打斷系統及非線性迴聲消除與主動式抗噪耳機

Abstract

本論文運用適應性演算法於語音打斷系統中回聲消除系統。並利用複合式回聲消除系統，提升回聲消除效率並藉此提升聲控裝置的辨識率。當使用者命令與互動裝置語音重疊時，能利用低運算量之偵測器避免回聲消除器發散。手持裝置上的放大器與微型喇叭可能會成非線性迴聲的產生，針對此種無法由線性回聲消除系統消除的成分，我們主要利用沃特拉濾波器(adaptive Volterra filter)及漢默斯坦模式(Hammerstein model)達到消除非線性迴聲的目的。本論文運用主動式噪音控制方式，利用音波重疊的原理，產生破壞性干涉的方法來消除噪音。 主動式噪音控制於低頻性能佳的優點，同時也不會造成體積龐大，亦可以彌補被 動式噪音控制的不足。

This work presents an acoustic echo canceller with barge-in handling capability aimed at voice-controlled devices. Barge-in refers to the situation in which an on-going machine vocal response is interrupted by the human’s command. Simultaneous voice rendering of the man and machine may degrade markedly speech recognition rate. An echo canceller comprising a fixed filter and an adaptive filter serves as a preprocessor to an automatic speech recognizer. A modified energy-based double-talk detector is incorporated into the echo canceller to ensure convergence of the adaptive filter. Through experiments, the proposed system shows improved cancellation performance and recognition rate compared to the conventional systems. The nonlinear distortions from low cost audio equipments degrade the performance of linear acoustic echo cancellation system. The approaches of nonlinear adaptive filter are have resorted to discrete-time transversal Volterra filters and block-based Hammerstein model. Active noise-reducing headset has been one of the most successful applications of active noise control. In this study, we focus on discussion of digital feedforward control.