- 學位論文
應用於耳機之內容認知的聲學音場增強系統
Context-Aware Sound Field Enhancement for Headphones
摘要
本論文所提出之內容感知的聲學音場增強(acoustic field enhancement)技術，以耳機為主，架構分為主副音分離、聲學音場增強和低音增強三大部分。在主副音分離的部分，透過商業用之音樂及電影之混音規則相關的數學特徵，進行二位元時頻遮蔽，分離出主音源：歌唱聲、對話聲，副音源：配樂伴奏聲及環境音等；聲學音場增強為針對不同的內容提供適當的增強方法，配樂伴奏聲的部分，透過音場寬度增強降低相關性方法，形成頭部音場外側化(externalization)，減低聽覺疲勞感，擴大感知音場，增加臨場感，在對話聲的部分，在總能量不超過配樂聲的範圍內，針對其子音頻帶透過參數型等化器(equalizer)予以增強，歌唱聲的部分則透過在主要的臨界頻帶(critical band)加入延遲脈衝、增加其空間感使其能夠和伴奏音場相匹配，又不失其主音源清晰度，最後再將加強完後之主音源和副音源合成，並考慮耳機之硬體特性，針對其低頻衰減處進行低音增強。實驗結果在電影的音訊檔部分，能夠有效的降低以往在聲學音場增強對主音源所造成的模糊感，總和上述三大架構，本論文之方法能夠帶給聽者絕佳的耳機聆聽感知。 關鍵詞：歌聲和伴奏聲分離，降低左右聲道相關性，頭部音場外側化，虛擬低音增強，語音清晰度增強
並列摘要
In this thesis, we proposed a technique of context-aware acoustic field enhancement under a listening condition using a stereo headphone. The architecture consists of three parts which are blind audio separation, acoustic field enhancement and virtual bass enhancement. In the blind audio separation system, we use commercial audio tracks’ mixture mathematical characteristics of left and right channels to carry out time-frequency binary mask, and then separate dominate sound (vocal and dialogues) and subordinate sound (incidental music and surrounding sound). In the acoustic field enhancement system, based on different audio property, we provide appropriate enhancers as follows: In the incidental music part, reduce left and right channels’ correlation to achieve externalization, decrease listening fatigue and expand the acoustic sound field as more as possible. In the speech part, we utilize an equalizer to enhance the consonants’ frequency bands, to achieve speech intelligibility. In the vocal part, the full search of two delay impulses at a time domain is conducted at difference critical bands, in order to make the vocal with the accompaniment more matchable. Finally, considering the physical characteristic of headphones, we adopt the missing fundamental phenomenon to enhance low frequency bands. The experimental results reveal that our method can make the dominate sound of movie audio more transparency compare with conventional acoustic field enhancement method. Summing of the above three systems, our method present excellent sound-field spaciousness, clear dialog, deep bass and immersive surrounding sense. Keywords：Blind Audio Separation, De-correlation, Externalization, Virtual Bass Enhancement, Speech Intelligibility Enhancement