201212659 六、發明說明： 【發明所屬之技術領域】 本發明為-種降低環境噪音之系統、方法 =，特別的是應用於麥歧陣列，透過即時的:二抑制程 序提供較好的通話品質的系統與其應用的裝置。 【先前技術】 利用通糾環㈣音造成的_ 1知技術提出 雙夕克風陣列（microphone array)降低 =，原理是設置-個接收語音與附近噪音的主麥兄克卞風，再 、個距離以外的位置設置另一個接收環境 =立兩個麥克風所接收的訊息透過計算可叫效消除環 兄呆曰，改善通話品質。 可：考第—圖顯不習知技術設置有兩個麥克風的通訊 署制專利公開第6,549,586號。其巾顯示的通訊 斑接麥克風，分別是遠離嘴巴的第—麥克風101 /巴，=第二麥克風102°第一麥克風101因為遠離嘴 語音.而作即是收集背景噪音’但也可能會收到通話201212659 VI. Description of the Invention: [Technical Field] The present invention is a system and method for reducing environmental noise =, in particular, applied to a Maiqi array, providing better call quality through an instant: two suppression program The system and the device to which it is applied. [Prior Art] Using the _ 1 knowing technology caused by the twisting ring (four) sound, the microphone array is lowered = the principle is to set a main microphone, which receives the voice and nearby noise, and then Set the other receiving environment in a location other than the distance = the message received by the two microphones can be used to eliminate the ringing of the ring and improve the call quality. Yes: The test-data is not known in the art. There are two microphones in the communication system. Patent No. 6,549,586. The communication display of the towel is connected to the microphone, which is the first microphone 101 / bar away from the mouth, = the second microphone 102 °, the first microphone 101 is the background noise because it is away from the mouth voice, but may also receive call
差二二帛二麥克風102即主要是收集通話語音，兩者的 ”可以作為抑㈣音的用途Q —麥克^中’為了先壓抑主要工作是收集背景噪音的第 體103後|〇1所收到的通話語音，訊號經過第一暫存記憶 背景噪:的由第—減法電路105處理降低通話語音來加強 話語部t計。相對地’由第二麥克風102所收集的通 104，其中^的背景噪音’訊號經暫存於第二暫存記憶體 八第二減法電路10ό先參考由第一減法電路1〇5經 4/24 201212659 時刻）所估計的背景噪音， 強抑制由第二麥克風102所 延遲電路107提供先前（前一 因此第二減法電路106可以加 收集的背景噪音訊號。 之後，有第三減法電路108同時接收第一減法電路奶 所估計的背景噪音與第二減法電路1G6所估計的語音訊 號’經參數調整後’可以得出_音抑制處理的語音财。 之後輸出至反傅立葉轉換電路（inverse fast Fol^erThe difference between the two microphones 102 is mainly to collect the voice of the call, and the two can be used as the use of the (four) tone Q - the microphone ^ in the first step to suppress the main work is to collect the background noise of the body 103 | The incoming call voice, the signal passes through the first temporary memory background noise: the reduced-to-speech voice is processed by the first-subtractive circuit 105 to enhance the utterance portion t. Relatively the 'pass 104 collected by the second microphone 102, where The background noise 'signal is temporarily stored in the second temporary memory 8 and the second subtraction circuit 10 first refers to the background noise estimated by the first subtraction circuit 1〇5 via 4/24 201212659), and is strongly suppressed by the second microphone 102. The delay circuit 107 provides the background noise signal that the previous subtraction circuit 106 can collect. The third subtraction circuit 108 simultaneously receives the background noise estimated by the first subtraction circuit milk and the second subtraction circuit 1G6. The estimated voice signal 'after parameter adjustment' can be used to obtain the voice of the _ tone suppression processing. Then output to the inverse Fourier transform circuit (inverse fast Fol^er
耐）109中，將離散時間的訊號轉變為連續的 頻域訊號，之後由疊加處理n 11G組合訊號，輸出語音訊 根據第-圖-般使用的麥克風陣列概念，之後更有習 知技術如Fortemedia™公司提出的難專利第7,587,〇56號 所揭露的麥克風陣列與抑制噪音的方法，其中進一步提出 更細節的處理流程，包括訊號調校（calibrati〇n)、波束形成 (beamforming)、噪音估計與壓抑、時域頻域轉換等的數 位處理方案’以求得到更好的通話品質。 然而’習知技術仍存在有一些缺點，比如： L由於缺乏有效調校的機制，所以對麥克風的品質要 求較高； ' 2. 使用固定式的波束成形（flxed beamforming)電路 k取語音訊號’將會要求麥克風之增益匹配； 3. 因為是固定式的波束成形技術，所以語音將會夹雜 較多的噪音，影響降噪的功能；如果想要進一步處 理降噪，則可能導致語音失真的問題。 【發明内容】 5/24 201212659 號調校求更好的通話品f ’本發明制利用即時的訊 適應性的波束形成技術與非線性的噪音抑制程 差，=1可以消除麥克風硬體差異或是設置位置產生的誤 炸=最大程度減少噪音，與提昇噪音抑制性能。、 有兩2貫施例’降低環㈣音U統主要是躺在—具 輸入端以上輸人端的收音模組中，此收音模組的 部ρI °又计用來主要接收語音或是特定音訊等主音訊 例，穿2要接收環境噪音部份’以本發明實現的裝置為 第一 有—麥克風陣列，麥克風陣酬至少包括-個 第麥克風模組與第二麥克風模組。 浦各ί克風模組所接收的訊號傳送至系統内部，訊號分 敏二各==克風模組收集聲音的靈 :對=:適應性波束二 :執部份相對極少的訊號。之後利用語音二 早^upil波，以此得出主音訊為主的訊號。 同時由“主的訊號與以環境噪音為主的訊後 產生用於抑制噪音的降噪增益，此增 資抽制的過_夠回饋到相關 6/24 201212659 根據貫施例，應用上述麥克風陣列的的嗓 則主要先接收麥克風陣列所收集的音訊，⑽=法 號判斷是轉有W卩份#環料音雜 = 增益，能用於調校目前音訊。 °u齐疋 經增益匹配的訊號接著執行波束形成的處理 ^員設門檻絲機·的效果，以有效得出環境噪音的 再將經難社音訊部份_縣音部份 Ϊ，利用另―預設門檻值調整濾波的效果，能夠梅取^ 音訊的部份。 只取於主 經轉換至頻域後，選擇性地進行訊 運算’調整到適當的訊號解析度 =與抽取 抑制的運算由上述頻域中的兩組 喿音 度，進而得出用於調整降噪的增益虎音的程 對連續的音訊在頻域巾執行降則 &降° 呆增益 【實施方式】 根據本發明實施例，在—實施例中， :=形成的麥克風陣列，目的是透過兩個收集不同 =置=日，㈣軟體或是硬體的實現，能 曰’進而得出品·好的音訊或是語音職。 根據本發明實現的技術，具有至少幾個優點· 1.1以有效抑制通話時的環妙音，而提高音喊祖 曰通訊過程中的清晰度與舒適度，不致被太多的環 具有兩個或兩個以上輸人端的收音模组中，^輸= 2 =個或以上的麥克風模組，收 7/24 201212659 境噪音所影響； 2. 本發明所提出的降低環境 訊號校準（caUbrat—，充=執，日守的 克風模組祕驗’可^忍各麥 〈刀貝）的增益差； 3. 此糸統中引入適應性波 . )技術摘取出一成:(：二 :提，大程度地減少音訊;爽雜的 利於b後端非線性噪音 — suppress)相關模組的性能。 /然本發明所提出的降低環境噪音除了㈣於特定 於具有兩個收音模組的裝置上，：仍 =麥克風形成的陣列’並非以本文所描述的實 實施例可參考第二圖。 塊圖圖=Γ降低環境嚼音之系統之模組化功能方 二=利用兩個輸入端的收音模組，用以接收一以 一以環境噪音為主的訊號，輸入端可 麥克風陣列:U組201與第二麥克風模組202形成的 接二二風模組201特別是設計用於收集欲 晋=、1L 权音訊，若用於通訊裝置上，可設 收集产产<Ρ ί的位置’第二麥克風模組202則是設計用於 &二H ’若用於通訊裝置上，可設置於離開第一麥 = ^01有—定距離的位置上，降低收集到語音訊號 或疋特定音訊的比例。 各麥克風模組所接收的訊號傳送至系統内部，本發明 201212659 所提出的线可以-龍電路（Ic)實現，或是透過軟體 手段實施。彡駄纯財相互紐賴的能即時線上調 校（“-time 0nline calibrati〇n)收音模組所接收的音訊的 調校早凡203、為能夠根據實際需要調整接收能量的波束妒 成單元綱、用於練主音糊_#音錄單元（辦砂 extraCt〇〇2G5、執彳怖_域域轉_躺轉 可變頻率解析轉換電路，variablefreq職cy_iuti〇nIn the 109, the discrete time signal is converted into a continuous frequency domain signal, and then the n 11G combined signal is processed by the superimposition, and the voice signal is output according to the concept of the microphone array used in the first picture, and then more conventional techniques such as Fortemedia The microphone array and noise suppression method disclosed in the difficult patent No. 7,587, 〇56, which is further proposed by the company, further proposes a more detailed processing flow, including signal calibration, beamforming, noise estimation. Digital processing schemes such as suppression, time domain frequency domain conversion, etc., in order to obtain better call quality. However, there are still some shortcomings in the conventional technology. For example: L lacks the effective tuning mechanism, so the quality of the microphone is high; ' 2. Use a fixed beamforming circuit to take the voice signal' The gain of the microphone will be required to be matched; 3. Because it is a fixed beamforming technique, the speech will be mixed with more noise, affecting the function of noise reduction; if you want to further deal with noise reduction, it may cause speech distortion. problem. [Summary of the Invention] 5/24 201212659 Tuning for better call products f 'The invention uses instant signal-adaptive beamforming technology and nonlinear noise suppression path difference, =1 can eliminate microphone hardware differences or It is the mis-explosion generated by setting the position = minimizing noise, and improving noise suppression performance. There are two or two examples of 'lowering ring (four) sound U system is mainly lying in the radio module with the input end and the input end, the part of the radio module is used to mainly receive voice or specific audio. In the case of the main audio, the wearer 2 is to receive the ambient noise portion. The device implemented by the present invention is the first microphone array, and the microphone array includes at least a first microphone module and a second microphone module. The signals received by the Puqiu gram module are transmitted to the inside of the system, and the signals are sensitive to each other == gram module collects the sound of the sound: pair =: adaptive beam 2: the relatively small number of signals. Then use the voice two early ^upil wave to get the main audio signal. At the same time, the noise reduction gain of the main signal and the environmental noise-based signal is used to suppress the noise. This capital increase is fed back to the relevant 6/24 201212659. According to the example, the above microphone array is applied. The 嗓 主要 接收 接收 接收 接收 接收 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风 麦克风The effect of performing beamforming is to set the effect of the door squeezing machine, so as to effectively obtain the ambient noise, and then use the other preset threshold value to adjust the filtering effect. The part of the audio signal is taken only after the main frequency is converted to the frequency domain, and the operation is selectively adjusted to the appropriate signal resolution = the operation of the decimation suppression by the two sets of frequency in the above frequency domain. And further, the process for adjusting the gain of the noise reduction is performed on the continuous tone in the frequency domain, and the drop is performed. [Embodiment] According to an embodiment of the present invention, in the embodiment, := Formed microphone array, It is through the two collections of different = set = day, (four) software or hardware implementation, can be 'goods' good audio or voice jobs. The technology implemented according to the invention has at least several advantages · 1.1 In order to effectively suppress the ring sound of the call, and improve the clarity and comfort during the communication process, it is not too many rings to have two or more input terminals, ^ lose = 2 = one or more microphone modules, which are affected by the noise of 7/24 201212659; 2. The environmental signal calibration proposed by the present invention (caUbrat-, charge=execution, the defensive structure of the defensive module) can be ^ The difference in gain of Niumeimai (knife); 3. The adaptive wave is introduced in this system.) The technology is extracted one by one: (: two: mention, greatly reduce the audio; Linear noise - suppresses the performance of the associated module. / However, the proposed reduction of ambient noise in addition to (d) on a device specific to two radio modules: still = the array formed by the microphone 'not described in this article For the embodiment, reference may be made to the second figure. The modular function of the system for reducing the ambient chewing sound = using two input audio modules to receive a signal based on ambient noise, the input microphone array: U group 201 and second microphone The second and second air modules 201 formed by the module 202 are specially designed to collect the audio signals of the desired and the 1L, and if used for the communication device, the second microphone module can be set to collect the position of the production <Ρ ί. 202 is designed for & two H' if used in the communication device, can be set at a distance from the first wheat = ^01, to reduce the proportion of collected voice signals or specific audio. The signal received by the module is transmitted to the inside of the system, and the line proposed by the present invention 201212659 can be implemented by the Dragon Circuit (Ic) or by software means.彡駄 财 财 相互 能 能 即时 即时 即时 即时 即时 “ “ “ “ “ “ “ “ “ “ “ “ “ “ “ 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 203 , used to practice the main sound paste _# audio recording unit (run sand extraCt 〇〇 2G5, 彳 彳 _ domain domain turn _ lying variable frequency analysis conversion circuit, variablefreq job cy_iuti〇n
触金麗，WRT) 、可執行雜輯音抑制的噪音抑 制早兀207、執仃頻域/時域訊號轉換的反頻域轉換如 反了變财解析㈣電路，—Se variable ftequeney resdut疆transformer ) 2〇8與執行職重疊加绝 (ovedap-add-sum)的疊加單元 2〇9。 〜 運作時，收音模組中的輸入端，如圖示中第 可設置於距離主要音訊來源較近的 集的聲音為音訊來源所產生的音訊；輸人端如第二麥= =且202則可設置於離開音訊來源稍遠的 ^集環境噪音射能會收細部份的音 要^ 麥克風模組加，搬產生的訊號分別標示為mi: =。各 訊細與M2分別傳送至調校單元加，調校單元2〇3 j於上歧音触，實際實現林核 的主音訊或環境噪音的 = 曰抽取早凡產生用於判斷當下 匕秸 !的資訊SF1，與噪音抑制單元產生用:判 疋否為環境噪音的資訊则） 、斷田下^虎 的靈敏度，能減低因為各麥克^且組收集聲音 設計的差昱、製造過彳。$& A 、， &差異（比如硬體 …錢過矛王產生的誤差、其中電路的差異等） 9/24 201212659 造成的誤差，在輸入端的調校可以確保之後訊號品質。上 述主音訊或環境噪音的資訊即為系統中後端元件所判斷 得出的訊號資訊（NF1與SF1)。Touch Jinli, WRT), perform noise suppression of noise suppression early 207, perform inverse frequency domain conversion of frequency domain/time domain signal conversion, such as inverse financial analysis (4) circuit, -Se variable ftequeney resdut transformer 2〇8 and the superposition unit 2〇9 of the ovedap-add-sum. ~ In operation, the input terminal in the radio module, as shown in the figure, can be set to the audio generated by the audio source in the set closer to the main audio source; the input end is the second wheat == and 202 It can be set to a little farther away from the source of the audio. The ambient noise can be the part of the sound that needs to be collected. The microphone module is added, and the signals generated by the movement are marked as mi: =. Each signal and M2 are respectively transmitted to the adjustment unit plus, and the adjustment unit 2〇3 j is in the upper ambiguous touch, and the actual realization of the main audio or environmental noise of the forest core = 曰 extraction is used to determine the current stalk! The information SF1, and the noise suppression unit are used to: determine whether it is the information of the environmental noise), and the sensitivity of the tigers under the field can reduce the difference in the sound design of each microphone and the collection. $& A,, & difference (such as the hardware ... the error caused by the money, the difference in the circuit, etc.) 9/24 201212659 The error caused by the adjustment at the input can ensure the quality of the signal afterwards. The above information about the main audio or ambient noise is the signal information (NF1 and SF1) judged by the back-end components in the system.
由調校單元203處理後，分別產生訊號81 (主音訊部 份），S2 (環境音訊部份），再分別傳輸至麵接於調校單元 203的波束形成單元（beamforming) 204。各麥克風模組接 收聲音的方向或角度代表所接收的訊號能量，為了獲得較 適當的接收能量，除了可以調整麥克風角度外，更能透過 波束形成的技術，針對由多個麥克風模組形成的麥克風陣 列所收集的聲波’聲波之間會相互干擾，產生干擾圖像 (mterference pattern )，對此依據設計需求調整成適當的干 擾圖像。 訊號經由波束形成單元204處理後，可產生語音訊; 或是特定音訊（主音訊部份）相對極少的職Ri，根據£ 示’代表主音訊部份的訊號S1與此主音訊極少的訊號r 同時傳輸至語音抽取單元2〇5。此語音抽取單元2G5主㈣ 行-滤波手段，經比對訊號S1與R1，輸出的訊號則^After being processed by the calibration unit 203, signals 81 (main audio portions) and S2 (environment audio portions) are respectively generated and transmitted to beamforming units 204 which are connected to the calibration unit 203, respectively. The direction or angle of the sound received by each microphone module represents the received signal energy. In order to obtain a proper receiving energy, in addition to adjusting the microphone angle, the beam forming technology can be used for the microphone formed by the plurality of microphone modules. The sound waves collected by the array 'sound waves will interfere with each other, creating a mterference pattern, which is adjusted to an appropriate interference image according to design requirements. After the signal is processed by the beam forming unit 204, a voice signal can be generated; or a relatively small job Ri of a specific audio (main audio portion), according to the signal indicating that the main audio portion of the signal S1 and the main audio signal is very small r At the same time, it is transmitted to the voice extraction unit 2〇5. The voice extraction unit 2G5 main (four) line-filtering means, after comparing the signals S1 and R1, the output signal is ^
作為前後訊號的調校參考，提㈣ 曰讯疋否主要有欲㈣的語音訊號或是特定音 士 SF1=0表示音訊主要為環境噪音；SH=i則表干 的語音訊號，特定音訊），輸出經過語音抽取的訊二U 頻域轉換早12〇6耗接於語音抽取 接 述主音訊部份相對極少的訊號Ri與經過語二= 號:1’也就是分別接收了主音訊部份與環境;= 。孔諕，以執仃一時域_頻域轉換程序 ， domain) rf f 汛戒由時域（tim< ;轉換為賴（frequeneydc)main)上，主要細 10/24 201212659 疋透過h夬速傅立葉轉換（Fast F〇urier Transforrnation)運算， • 分別產生頻域訊號P1與P2。 接著，噪音抑制單元207由頻域轉換單元200接收頻 域訊號P1與P2,藉此估計出環境噪音，並產生用於降低噪 音的增益（gam)，即訊號G1。另產生用於表達當下訊號是 否為噪音的訊號NF1，回饋至調校單元2〇3作為前後訊號 的調校參考。 反頻域轉換單元20δ接收增益訊號G1與頻域訊號 鲁 P1，利用增益訊號G1作為内插法的依據，與訊號ρι運算 反! 夬速傅立葉轉換（inverse past Fourier Transf〇rmatj〇n)， 將頻域§fl號轉換為時域中，產生時域訊號。 最後，各時段的訊號SOI將經疊加單元209加總，形 成一連續輸出的音訊。 上述各模組單元的運作細節可接著參考以下各圖示與 流程，當中各模組當可以軟體手段或是硬體電路實現。μ 第二圖顯不為本發明降低環境噪音的系統中調校單元 # 2 〇 3實施例之一的工作流程圖，其中系統先接收由上述第一 麥克風模組201與第二麥克風模組2〇2產生的訊號組盘 $，在此實施例中，M1為主要為主音訊部份的訊號，通 常同時包括有語音訊號與環境噪音，而M2則主要為環境噪 音的訊號，但仍會包括部份語音訊號。 〃 先如步驟S301，訊號SF1係為上述語音抽取單元2〇5 所產生的訊號，訊號經語音抽取的程序，透過訊號則表 達所含之tfl制容，若SFl=〇，麵音财要為環境噪^ 步驟將直接進入步驟S31G ;若SF1 =卜則表示訊號 /、有欲收集的語音職或是特定音訊（是），訊息將帶入步 11/24 201212659 作為累加次數 驟S303，作為計算參考，並帶入步驟S304 , 的參考。 相對地，如步驟S302，訊號NF1為上述。喿音抑帝μ 一 207產生回饋的訊號，以此判斷訊號是否為環境噪音早70 NF1=0 (否）’也就是判斷當下的訊號為主音訊部::步= 到S310進行整合；若NFi=i (是），訊息將帶入步驟， 以此為參考計算功率，或帶入步驟S305，作為累加環产4 音次數的參考。 τ兄木 如步驟S303，計算訊號Ml與M2的功率（能量），、、， 分別產生能量Pol與P〇2，能量Pol與p〇2將分別作為執 行如步驟S304與S305中訊號判斷步驟的參考。 在步驟S304中，若接收的訊號為語音訊號或是特定音 訊，將累加次數（Cntl)，若累加次數尚未超過_門檻二 (Cntl<Thl)(否）’先執行步驟S31〇 ;若累加次數超過這 門檻值（是）’則於步驟S308中計算增益（Gainl)。 上述訊號Pol與P〇2中的主音訊部份經能量累加（步 驟S304)後，記載於訊號SS1與SS2中。當步驟§3〇6 $ 斷主音訊部份的次數（Cntl )已超過特定門檻 (Cntl Thl ) ’上述經累加的訊號ssi與μ]則會於步驟 S308計算後得出增益Gaini。 另一方面’圖示右方的流程將處理環境噪音的部份， 經計算出的功率值P〇l與P〇2，匯同訊號NF1所帶的資訊 (NF1=0表示非環境噪音；Νπ=1表示為環境噪音），於步 驟S3〇5累加為環境噪音的訊號次數（Cnt2)，並累加環境 噪音的功率（能量），訊息載於訊號SN1與SN2。 當累加的環境噪音次數達到一門檻（Cnt2>=Th2)時， 12/24 201212659 步驟將進入S309，由訊號SN1與SN2所載的資訊計管垆只 (叫產生增益累加的環境嚼音訊號 未達到門檻（Cnt2<Th2) ’步驟將直接至S3l〇處理。 步驟S310是融合（gainfusion)各增益Gainl與Gain2, 並參考自訊號SF1與NF1所載主音訊部份或環境噪音部份 的資訊’進行整合得出增益Gain。其中決定增益咖的方 式可有多種，其中之-是：由於音訊不斷地進人此系統中， ㈣步驟S31G僅獲得Gainl的資訊，有時就僅有純2的 貧訊，若是沒有Gainl與Gain2，則增益Gain為丨。除了第 二圖所描述的程序與判斷外，各增益Gain，Gainl，Gain2的 计异為一般技術，為熟悉此項技術之技術人員可據以得出。 隶後，如步驟S311，增盈Gain將施加於第一麥克風模 組產生的訊號Ml與第二麥克風模組產生的訊號M2上，分 別輸出經增益調整後的訊號S1與S2。 第四圖則接著顯示本發明系統中波束形成單元2〇4之 板組化功此方塊圖，圖中顯示的各單元方塊可以軟體手段 達成，或是可以硬體電路實現。 圖中顯示的波束形成單元204接收經增益調整的訊號 S1與S2’先經過功率計算單元401分別計算出訊號功率（能 量）’產生訊號PS 1與PS2。接著透過語音偵測單元4〇3偵 測訊號中的主音訊部份，包括語音部份、特定音訊等。根 據實施例’可先透過語音偵測單元403判斷PS1與PS2的 能量差異是否大於預設門檻（第一預設門播），根據此門檻 決疋參數VI ’以此參數VI控制濾波單元405中的濾、波係 數（filter coefficient)。透過一個延遲單元407延遲訊號si 與直接進入濾波單元405的訊號S2，經此濾波手段產生語 13/24 201212659 音訊號較少的訊號R1。 之後，可參考第五圖顯示系統中語音抽取單元2〇5之 模組化功能方塊圖，語音抽取單元205同樣可以軟體手段 達成’或是可以硬體電路實現。 語音抽取單元205接收語音訊號較少的訊號R1 (主要 為環境噪音）與之前經增益調整後的訊號S1 (主要為主音 訊部份）’同樣先透過功率計算單元5〇1計算出個別的功 率，產生訊號PS1 (已由波束形成單元204中的功率計算單 兀401產生）與prj，透過語音偵測單元5〇3酬兩個能 量的差異是否大於另-預設門檻（第二預糾插），以此根 據產生參數V2 ’用以控制濾波單元5〇5中的濾波係數，能 夠產生適應性（adaptive )的遽波效果。圖示中濾波單元5〇5 同時接收經延遲單元5〇9延遲的訊號S1與訊號ri，藉此 產生具有主音訊部份的訊號，也就是主要為語音訊號座 特定音訊的輸出訊號。 、 語，取單* 2〇5 t具有一個語音確認（― _麵）早凡507,語音確認單元5〇7榻取訊號PS1盥PR1， ^其中，出此時通過的職衫絲音訊號或特定音 = 1 ;反之，設定.訊號 。 考' 回胃至_單元2G3作為雛麥歧訊號的參 擦组化二考第'、圖所不系統中頻域轉換單元206之 訊號較少=2G6 _號A1與語音 體電路。制的是，軟料段或是硬 元6〇1與6G3進行快速^與R1分別透過傅立葉轉換單 、、傅立葉轉換，產生的頻域訊號為FA1 14/24 201212659 與FRl，並且進行訊號轉換時，可透過取樣的 機制降低計算量。頻域訊號為FA1與FR1，接著可繼浐八 ，經過平滑與抽取單元6Q2, _執行平滑（迎⑽㈣）運 算與抽取（dedmating)運算，能在不失真的航下刪除干 擾的訊號、運作較少的訊號降低運算成本，能優化訊號處 理流程。—，此程序為選擇性，並非必要。最後分別= 生訊號P1與P2。 經頻域轉換後的訊號P1與p2傳遞至噪音抑制單元As a reference for the adjustment of the front and rear signals, mention (4) whether the main voice (4) or the specific SF1 = 0 indicates that the audio is mainly ambient noise; SH = i is the dry voice signal, specific audio), The output is subjected to voice extraction. The second U frequency domain conversion is earlier than 12 〇6. The speech is extracted. The main audio part is relatively rare. The signal Ri and the utterance two = number: 1', that is, the main audio part is received separately. Environment; = . Kong Wei, to perform a time domain _ frequency domain conversion program, domain) rf f 汛 ring from the time domain (tim <; converted to 赖 (frequeneydc) main), the main fine 10/24 201212659 疋 through h 傅 Fourier transform (Fast F〇urier Transforrnation) operation, • Generate frequency domain signals P1 and P2, respectively. Next, the noise suppression unit 207 receives the frequency domain signals P1 and P2 from the frequency domain converting unit 200, thereby estimating the environmental noise, and generating a gain (gam) for reducing the noise, that is, the signal G1. Another signal NF1 for expressing whether the current signal is noise or not is fed back to the calibration unit 2〇3 as a calibration reference for the front and rear signals. The inverse frequency domain conversion unit 20δ receives the gain signal G1 and the frequency domain signal Lu P1, and uses the gain signal G1 as the basis of the interpolation method, and operates inversely with the signal ρι! Inverse past Fourier Transf〇rmatj〇n, The frequency domain §fl number is converted into the time domain to generate a time domain signal. Finally, the signal SOI for each period will be summed by the superimposing unit 209 to form a continuously outputted audio. The details of the operation of each of the above module units can be followed by the following diagrams and flows, where each module can be implemented by software or hardware. The second figure is not a working flow chart of one of the embodiments of the calibration unit #2 〇3 in the system for reducing environmental noise, wherein the system first receives the first microphone module 201 and the second microphone module 2 In the present embodiment, M1 is the signal mainly for the main audio portion, which usually includes both voice signals and ambient noise, while M2 is mainly for ambient noise signals, but will still include Part of the voice signal. 〃 First, in step S301, the signal SF1 is a signal generated by the voice extraction unit 2〇5, and the signal is subjected to a voice extraction process, and the signal is expressed by the tfl system. If SFl=〇, the surface audio is The environmental noise step will go directly to step S31G; if SF1 = Bu, the signal /, the voice job to be collected or the specific audio (yes), the message will be taken to step 11/24 201212659 as the accumulation number S303, as the calculation Refer to and bring to the reference of step S304. In contrast, as in step S302, the signal NF1 is as described above.喿音抑帝 μ 207 generates a feedback signal to determine whether the signal is ambient noise early 70 NF1=0 (No) 'That is to judge the current signal as the main audio department:: step = to S310 for integration; if NFi =i (Yes), the message will be taken to the step, and the power is calculated as a reference, or taken to step S305 as a reference for accumulating the number of 4 tones of the ring. τ 兄木, as in step S303, calculates the power (energy) of the signals M1 and M2, and generates energy Pol and P 〇 2, respectively, and the energies Pol and p 〇 2 will respectively be used as steps for performing the signal determination in steps S304 and S305, respectively. reference. In step S304, if the received signal is a voice signal or a specific audio, the number of times (Cntl) will be accumulated. If the accumulated number of times has not exceeded _ threshold 2 (Cntl < Thl) (No), step S31 is performed first; if the number of accumulations is If the threshold value is exceeded (Yes), the gain (Gainl) is calculated in step S308. The main audio portions of the signals Pol and P〇2 are accumulated by energy (step S304), and are recorded in the signals SS1 and SS2. When the number of times of the main audio portion (Cntl) of step §3〇6 $ has exceeded a certain threshold (Cntl Thl ) 'the above accumulated signals ssi and μ', the gain Gaini is obtained after calculation in step S308. On the other hand, the process on the right side of the diagram will process the part of the ambient noise, and the calculated power values P〇l and P〇2, together with the information carried by the signal NF1 (NF1=0 means non-environmental noise; Νπ =1 indicates ambient noise), the number of signal noises (Cnt2) accumulated in the ambient noise in step S3〇5, and the power (energy) of the ambient noise is accumulated, and the message is carried on the signals SN1 and SN2. When the accumulated number of ambient noise reaches a threshold (Cnt2>=Th2), the 12/24 201212659 step will proceed to S309, and the information contained in the signals SN1 and SN2 will only be called (the environment is called the gain accumulating environment. Steps to reach the threshold (Cnt2 <Th2) will be directly processed to S3l. Step S310 is to fuse the gains Gainl and Gain2, and refer to the information of the main audio part or the ambient noise part contained in the signals SF1 and NF1. Integration can be used to derive Gain. There are many ways to determine the gain of the coffee, among which: because the audio is continuously entered into the system, (4) Step S31G only obtains Gainl information, sometimes only pure 2 is poor. If there is no Gainl and Gain2, the gain Gain is 丨. In addition to the procedures and judgments described in the second figure, the Gain, Gainl, and Gain2 gains are general techniques, and can be used by those skilled in the art. After the step S311, the gain Gain is applied to the signal M1 generated by the first microphone module and the signal M2 generated by the second microphone module, and the gain-adjusted signals S1 and S are respectively output. 2. The fourth figure then shows the block diagram of the beam forming unit 2〇4 in the system of the present invention, and the unit blocks shown in the figure can be achieved by software means or can be implemented by a hardware circuit. The beam forming unit 204 receives the gain-adjusted signals S1 and S2', respectively, and calculates the signal power (energy) by the power calculating unit 401 to generate signals PS 1 and PS2. Then, the signal is detected by the voice detecting unit 4〇3. The main audio portion includes a voice portion, a specific audio, etc. According to the embodiment, the voice detection unit 403 can first determine whether the energy difference between PS1 and PS2 is greater than a preset threshold (the first preset gate broadcast). The threshold parameter VI' controls the filter coefficient in the filtering unit 405 by this parameter VI. The delay signal 407 delays the signal si and the signal S2 directly entering the filtering unit 405, and the filtering means generates the language 13 /24 201212659 Signal R1 with less audio signal. After that, refer to the fifth figure to display the modular function block diagram of voice extraction unit 2〇5 in the system, voice extraction list. The element 205 can also be implemented by software means or can be implemented by a hardware circuit. The voice extraction unit 205 receives the signal R1 (mainly ambient noise) with less voice signal and the previously adjusted signal S1 (mainly the main audio part). The same is to first calculate the individual power through the power calculation unit 5〇1, generate the signal PS1 (which has been generated by the power calculation unit 401 in the beam forming unit 204) and prj, and transmit the two through the voice detection unit 5〇3 Whether the difference in energy is greater than the other-preset threshold (second pre-interpolation), thereby generating an adaptive chopping effect according to the generation parameter V2' for controlling the filter coefficients in the filtering unit 5〇5. In the figure, the filtering unit 5〇5 simultaneously receives the signal S1 and the signal ri delayed by the delay unit 5〇9, thereby generating a signal having a main audio portion, that is, an output signal mainly for the audio signal. , language, order form * 2〇5 t has a voice confirmation (― _ face) early 507, voice confirmation unit 5 〇 7 couch take signal PS1 盥 PR1, ^ which, through the shirt thread signal or Specific tone = 1; otherwise, set the signal. Test 'Back to the stomach to _ unit 2G3 as the brooding error number of the two chapters, the signal in the system is less in the frequency domain conversion unit 206 less = 2G6 _ number A1 and the voice body circuit. The system is that the soft segment or the hard cell 6〇1 and 6G3 are fast and R1 are respectively transmitted through the Fourier transform single and Fourier transform, and the generated frequency domain signals are FA1 14/24 201212659 and FR1, and the signal conversion is performed. The calculation can be reduced by the sampling mechanism. The frequency domain signals are FA1 and FR1, and then can be followed by eight. After smoothing and extracting unit 6Q2, _ performs smoothing (welcome (10) (four)) operation and decimation (dedmating) operation, which can delete the interference signal and operate after undistorted navigation. Less signal reduces computational cost and optimizes signal processing. — This procedure is optional and not necessary. Finally, respectively = raw signals P1 and P2. The frequency-domain converted signals P1 and p2 are transmitted to the noise suppression unit
207 ’可參考第七圖所示系統中噪音抑制單元2〇7之模组化 功能方塊圖。 '' 噪音抑制單it 2〇7同樣可為軟體手段達成，或是可以 硬體電路實現，在本發明實關，此為後段的噪音抑制手 段，可以勿、略。 。呆音估計單7〇 701主要是執行非線性噪音抑制程序 /nonlinear noise suppressi〇n)，能夠根據訊號卩1與打估 計出環境噪音’並計算得出訊號調整用的增益GQ，同時產 生訊號NF卜也就是輸人至調校單元朋巾的參考訊號， 以此表示該段訊號是否主要為環境噪音，比如若為環境啤 音’可設定NF卜1 ;若為主音訊部份，則設定Νπ=〇。產 生的增益GG可再經增益校正單元7()3處理，輸出用於 用的增益G1。 …增益G1之後傳遞至反頻域轉換單元通，反頻域轉換 單7L 208再接收上述載有主音訊部份的訊號W，根據辦兴 I1進行調整’達成降細目的。反頻域轉換單it施内^ 貫現可參考第八圖所*的模組化功能方塊圖。 經非線性噪音㈣触射生的增益G1將可有效抑 15/24 201212659 制主曰5孔部份的π呆音，增益訊號G1先經内插（jnteip〇iati〇n ) 單元801調整回時域中的增益ιοί，與訊號p]逐點對應相 乘，產生頻域訊號GP1，最後經反傅立葉轉換單元803轉 換回時域的訊號，輸出訊號S01。 疊加單元209耦接於反頻域轉換單元2〇8,接收其輸 出的訊號soi ’此訊號在時域中以波形表示，疊加單元2〇9 將聲波經重疊（Qvedapping)、相加（adding)與訊號加總 (summing)等運算形成連續的音訊輸出。207 ' can refer to the modular function block diagram of the noise suppression unit 2〇7 in the system shown in the seventh figure. '' The noise suppression single IT 2〇7 can also be achieved by software means, or can be implemented by a hardware circuit. In the present invention, this is a noise suppression method in the latter stage, which can be omitted. . The overtone estimate 7〇701 is mainly to perform a nonlinear noise suppression program/nonlinear noise suppressor(〇), which can estimate the ambient noise according to the signal 卩1 and calculate the gain GQ for signal adjustment, and generate the signal NF at the same time. Bu is also the reference signal for the input to the adjustment unit, to indicate whether the signal is mainly environmental noise. For example, if the environment is a beer sound, you can set NF Bu 1; if it is the main audio part, set Νπ =〇. The generated gain GG can be processed by the gain correcting unit 7() 3 to output a gain G1 for use. The gain G1 is then passed to the inverse frequency domain conversion unit, and the inverse frequency domain conversion unit 7L 208 receives the signal W carrying the main audio portion, and adjusts according to the commercial I1 to achieve the purpose of thinning. The inverse frequency domain conversion single-input can be referred to the modular function block diagram of the eighth figure. The gain G1 through the nonlinear noise (4) contact can effectively suppress the π dullness of the 5 hole portion of the 15/24 201212659 master, and the gain signal G1 is first adjusted by interpolation (jnteip〇iati〇n) unit 801. The gain ιοί in the domain is multiplied by the signal p] point by point to generate the frequency domain signal GP1, and finally converted back to the time domain signal by the inverse Fourier transform unit 803, and the signal S01 is output. The superimposing unit 209 is coupled to the inverse frequency domain converting unit 2〇8, and receives the signal SOO′ of the output thereof. The signal is represented by a waveform in the time domain, and the superimposing unit 2〇9 overlaps (Qvedapping) and adds the sound wave. Computation with signals, such as summing, forms a continuous audio output.
經上述各電路模組，本發明所應用的方法則歸納為第 九圖所示為應用本發明提出的降低環境噪音之纟統所執行 的降低環境噪音的方法流程。 y很料發明實_，上述各功能方塊可讀體手段 :哭：序：程式化於一内嵌晶片中，或是可載入系統中 理為的記憶體中。 τ 夕克風陣列中至少具有一個主Through the above circuit modules, the method applied by the present invention is summarized as the flow chart of the method for reducing environmental noise performed by the system for reducing ambient noise proposed by the present invention. y is expected to be invented _, the above functional block readable means: cry: order: stylized in an embedded chip, or can be loaded into the memory of the system. There is at least one master in the τ 克 风 wind array
的降低環境噪音的流程所部發明系統所執 至少如麥克風陣列)收集音訊後(步驟S901 執;增益匹配。利用調二 訊號判斷是否= 主要是包括系統根據之 (步驟S9〇3)，藉以決是環境噪音部份的資， 音訊’即麥克風陣列所接二音 16/24 201212659 序，1::二ΐ::5)。此調校過程為持續進行的程 話品質。㈣麥克風與環境的狀況，提供較佳的通 要是配的訊號接著執行波束形成的處理程序，主 個二比如判斷兩 值的狀況來』==異疋否超過-個預物 (步驟以有效得出環境噪音的部份 吟音方法步驟S9G7利用上述得出的環境 曰π (主曰δίι部份相對極少的訊號），藉以與第—麥 ί風模組得出且經過調校社音訊部份的訊號比對，1差 ::再與另一預設門檻值比較’用來調整遽波的效果， 月b夠擷取於主音訊的部份（步驟S909)。 透過上述步驟S905與步驟S9G7分別得出環境噪 伤與主音訊部份’接著執行時域_頻域轉換（步驟s9u)， 比如利用快速傅立葉轉換程序將訊號於時域中轉換為頻域 上的訊號，可再選雜地進行訊號平滑運算與抽取運算， 調整到適當㈣_析度，最後制㈣加程序還原訊 號，用適當節省的運算資源產生好的通話品質。 經時域-頻域轉換後，利用一種非線性噪音抑制的運算 由上述頻域中的兩組訊號估言十出環境噪$的程度（步驟 S913)，進而得出用於調整降噪的降噪增益（步驟沾15)。 最後利用此降°呆增益對連續的音訊在頻域中執行降噪 (步驟S9n) ’再轉換為時域訊號，如應用反快速傅立葉轉 換（步驟S919)，最後再經訊號重疊與加總流程後輪 驟 S921)。 17/24 201212659 兩個音的系統與其方法則特別應用於具有 中透過訊賴校、波束===境噪音之系統，其 噪音抑制與疊加的程序後 $頻域/時域轉換、 ::號進行即時處理，隨時根;情:=麥二風輪出 抑制通話_環境噪音，而 ’可以有效 清麥克， _本發明之專：發::=:，，非因此即 内容所為之等效έ士構’卜％運用本毛明說明書及圖示 内，合予陳明 《化，均同理包含於本發明之範圍 【圖式簡單說明】 第一圖顯* f知技術雙麥克風通訊裝 第二圖顯示本發明降低产士”二…:置電路方塊圖’· 方塊圖；科辑私㈣音之系統之模纽化功能 •苐二圖顯示為本發H統中調校單 作流程圖； 元之實施例之一 工 能方=圖顯示為本發明系統中波束形成單元之模組化功 能方：圖顯示為本發明系統中語音抽取單元之模組化功 元之模組化功 能方^圖顯示為本發料統中頻域轉換單元之模組化功 第七圖顯示為本發明系統中噪音抑制單 <8/24 201212659 能方塊圖； 帛人ffi顯7F為本發明系統巾反賴轉換單元之模組化 功能方塊圖； 第九圖顯示為應用本發明系統所執行的降低環境噪音 的方法。 【主要元件符號說明】 第二麥克風102 第二暫存記憶體104 第二減法電路1〇6 第三減法電路108 疊加處理器110The process of reducing the environmental noise is performed by the invention system at least as the microphone array collects the audio (step S901 is performed; the gain is matched. The second signal is used to determine whether = mainly includes the system according to (step S9〇3), It is the part of the ambient noise. The audio 'that is the microphone array is connected to the second tone 16/24 201212659, 1: 1:2::5). This tuning process is an ongoing process quality. (4) The condition of the microphone and the environment, providing a better general purpose is to match the signal and then perform the beamforming processing procedure, the main two, for example, to judge the condition of the two values ??? == 疋 超过 no more than - a pre-object (step to effectively Part of the noise method for the ambient noise step S9G7 uses the above-mentioned environment 曰π (the main 曰δίι part of the relatively few signals), and the audio-visual part of the tuned company The signal comparison, 1 difference:: and then compared with another preset threshold value' is used to adjust the effect of the chopping, and the month b is enough to capture the part of the main audio (step S909). Through the above steps S905 and S9G7 Obtaining the environmental noise and the main audio portion respectively, and then performing the time domain_frequency domain conversion (step s9u), for example, using the fast Fourier transform program to convert the signal in the time domain into a signal in the frequency domain, which can be selected again. The signal smoothing operation and the decimation operation are adjusted to the appropriate (four)_disaggregation, and the final system (4) plus program restores the signal, and uses the appropriate saved computing resources to generate good call quality. After the time domain-frequency domain conversion, a nonlinear noise is used. The operation is estimated by the two sets of signals in the above frequency domain to determine the degree of environmental noise $ (step S913), and then the noise reduction gain for adjusting the noise reduction is obtained (step 15). Finally, the drop gain is used. Noise reduction is performed on the continuous audio in the frequency domain (step S9n) 'reconverted to the time domain signal, such as applying inverse fast Fourier transform (step S919), and finally by signal overlap and summation flow rear wheel S921). 17/24 201212659 Two-tone system and its method are especially applied to systems with medium-passing, beam===land noise, noise suppression and superimposition procedures after frequency/time domain conversion, :: Instant processing, at any time; love: = Mai two winds out to suppress the call _ environmental noise, and 'can effectively clear the microphone, _ the specialization of the invention: send::=:,, therefore, the content is equivalent士建's use of this manual and the illustrations, and the combination of Chen Ming, all are included in the scope of the present invention [simplified description of the drawings] The first picture shows * f know the technology of dual microphone communication equipment The second figure shows that the invention reduces the maternal "two...: the circuit block diagram" · the block diagram; the system of the private (four) sound system of the model new function; the second picture shows the flow chart of the adjustment of the current unified system One of the embodiments of the element is shown in the figure. The figure shows the modular function of the beam forming unit in the system of the present invention: the figure shows the modular function of the modular power element of the voice extraction unit in the system of the present invention. ^The figure shows the modularization of the frequency domain conversion unit in this system. Figure 7 shows the noise suppression single <8/24 201212659 energy block diagram of the system of the present invention; 帛人 ffi显 7F is a modular function block diagram of the system towel smashing conversion unit of the present invention; Method for reducing environmental noise performed by the invention system [Description of main component symbols] Second microphone 102 Second temporary storage memory 104 Second subtraction circuit 1〇6 Third subtraction circuit 108 Superimposing processor 110
第一麥克風101 第一暫存記憶體103 第一減法電路105 延遲電路107 反傅立葉轉換電路1〇9 Λ號 Ml，M2，SI, S2, Rl，Al，SF1，Pl，P2, NF1，G1， SOI, Pol, P〇2, SSI, SS2, SN1, SN2, PS1, PS2, PR1, Al, FAl! FR1, GP1First microphone 101 first temporary memory 103 first subtraction circuit 105 delay circuit 107 inverse Fourier transform circuit 1〇9 M M1, M2, SI, S2, Rl, Al, SF1, Pl, P2, NF1, G1, SOI, Pol, P〇2, SSI, SS2, SN1, SN2, PS1, PS2, PR1, Al, FAl! FR1, GP1
參數V1，V2 第一麥克風模組201 調校單元203 語音抽取單元2〇5 。喿音抑制單元207 疊加單元209 功率計算單元401，501 濾波單元405,505 功率計算單元501 傅立葉轉換單元601，603 噪音估計單元701 增益 GO, Gl，IG1 第二麥克風模組202 波束形成單元204 頻域轉換單元206 反頻域轉換單元208 增益 Gain, Gain 1，Gain2 語音偵測單元403,503 延遲單元407,509 語音確認單元507 平滑與抽取單元602,604 增益校正單元703 19/24 201212659 内插單元801 步驟S301〜S311 步驟 S901~S921 反傅立葉轉換單元803 調校單元之工作流程 降噪流程 20/24Parameter V1, V2 First microphone module 201 Tuning unit 203 Voice extraction unit 2〇5. Arpeggio suppression unit 207 Superposition unit 209 Power calculation unit 401, 501 Filter unit 405, 505 Power calculation unit 501 Fourier conversion unit 601, 603 Noise estimation unit 701 Gain GO, Gl, IG1 Second microphone module 202 Beamforming unit 204 Frequency domain conversion Unit 206 inverse frequency domain conversion unit 208 gain Gain, Gain 1, Gain2 voice detection unit 403, 503 delay unit 407, 509 voice confirmation unit 507 smoothing and extraction unit 602, 604 gain correction unit 703 19/24 201212659 interpolation unit 801 steps S301 to S311 step S901 ~S921 Anti-Fourier Transform Unit 803 Calibration Unit Workflow Noise Reduction Process 20/24