201240423 六、發明說明： 【發明所屬之技術領域】 [0001] 本發明涉及一種雜訊去除系統及方法，特別涉及一種通 話雜訊去除系統及方法。 【先前技術】 [0002] 目前，利用手機、電話機等通話裝置將本方的聲音傳輸 至對方所使用的技術一般是：透過電磁感應將本方發出 的聲音轉換為類比訊號，經模/數轉換器將該類比訊號轉 化為數位訊號，再將該數位訊號放大後傳輸至對方。但 是在通話過程中，我們常常會遇到這種情況：在一個非 常嘈雜的環境中用手機或電話機與人通話，即使我們大 聲嚷嚷，對方卻仍然聽不清我們在說什麼。這是因為我 們說話的聲音中夾雜著環境雜訊，被同時傳輸給了對方 。顯然，這樣會嚴重影響雙方的通話品質。 【發明内容】 [0003] 鑒於以上内容，有必要提供一種通話雜訊去除系統及方 法，可以去除通話過程中環境帶來的雜訊，使用戶的聲 音清晰地傳送至對方。 [0004] 一種通話雜訊去除系統，運行於通話裝置中，該通話裝 置還包括語音識別系統和聲紋識別系統，該通話雜訊去 除系統包括：觸發模組，用於當通話裝置接通電話時， 觸發語音識別系統和聲紋識別系統開啟；所述觸發模組 還用於當偵測到用戶說出預先設置的開始指令時，觸發 聲紋識別系統開始提取通話裝置的話筒所採集聲音的聲 紋特徵；所述觸發模組還用於當偵測到用戶說出預先設 100110007 表單編號A0101 第4頁/共17頁 1002016889-0 201240423 置的結束指令時，觸發聲紋識別系統根據所提取的聲紋 特徵建立-個背景模型，以及觸發聲紋識別系統重新即 日後取所述話筒所採集聲音的聲紋特徵，並即時根據所 提取的聲紋特徵建立聲紋模型；比對模组，用於即時將 該聲紋模型與所述背景模型進行比對，得出該聲紋模型 中與月景模型-致和不—致的部分；所述觸發模組還用 於控制該聲紋模型中與該背景模型—致的部分對應的聲 音傳輸至對方。 [0005] ❹ [0006] 種通話雜讯去除方法，應用於通話裝置中，該通話裝 置包括語音識別系統和聲紋識別系統，該方法包括以下 步驟.（a)當通話裝置接通電話時，觸發語音識別系統 和聲紋識別系統開啟；（b)當偵測到用戶說出預先設置 的開始指令時’觸發聲紋識別系統開始提取通話裝置的 話筒所採集聲音的聲紋特徵；（c)當偵測到用戶說出預 先叹置的結束指令時，觸發聲紋識別系統根據所提取的 聲紋特徵建立一個背景模型；（d)觸發聲紋識別系統重 新即時提取所述話筒所採集聲音的聲k特徵，並即時根 據所提取的聲紋特徵建立聲紋模型；（e)即時將該聲紋 模型與所述背景模型進行比對，得出該聲紋模型中與背 景模型一致和不一致的部分，並控制該聲紋模型中與該 背景模型一致的部分對應的聲音傳輸至對方。 相較於習知技術，所述通話雜訊去除系統及方法，可以 去除通話過程中環境帶來的雜訊，使用戶的聲音清晰地 傳送至對方。 【實施方式】 100110007 表單編號A0101 第5頁/共17頁 1002016889-0 201240423 [0007] 參閱圖1所示，係本發明通話雜訊去除系統較佳實施方式 的運行環境圖。在本實施方式中，通話雜訊去除系統10 運行於通話裝置1中。該通話裝置1可以為手機、電話機 等。 [0008] 該通話裝置1中還包括話筒11、語音識別系統1 2、聲紋識 別系統13和儲存器14。其中，話筒11用於當通話裝置1接 通電話時，採集周圍環境的聲音，包括用戶使用通話裝 置1通話時的說話聲，以及環境的雜訊。 [0009] 語音識別系統1 2用於即時從話筒11所採集的聲音中識別 用戶說話的内容。聲紋識別系統13用於當用戶說出所設 置的開始指令，如“喂”時，即時提取話筒11所採集聲 音的聲紋特徵，以及當用戶說出所設置的結束指令，如 “你好”時，根據所提取的聲紋特徵建立背景模型。聲 紋識別系統13可以採用將所提取的聲紋特徵轉換為N維特 徵向量的方法建立背景模型。 [0010] 聲紋識別系統13還用於當用戶說出所述結束指令後，重 新即時提取話筒11所採集聲音的聲紋特徵並即時根據該 聲紋特徵建立聲紋模型。 [0011] 通話雜訊去除系統10用於即時將該聲紋模型與所述背景 模型進行比對，控制該聲紋模型中與該背景模型一致的 部分對應的聲音透過電磁感應和模/數轉換器作用後傳輸 至對方，從而將該聲紋模型中與該背景模型不一致的部 分對應的聲音，即環境的雜訊過濾。 [0012] 參閱圖2所示，係本發明通話雜訊去除系統較佳實施方式 100110007 表單編號A0101 第6頁/共17頁 1002016889-0 201240423 的功能模組圖。該通話雜訊去除系統10包括設置模組101 、偵測模組102、觸發模組103、獲取模組104和比對模 組 1 0 5。 [0013] 設置模組101用於設置開始指令和結束指令，並將該開始 指令和結束指令存入儲存器14中。該開始指令和結束指 令可以為用戶通話時常用的起始用語，如開始指令可以 為“喂”，而結束指令可以為“你好”等詞條· [0014] Ο [0015] 偵測模組102用於偵測通話裝置1是否接通電話，以及當 通話裝置1接通電話後是否掛斷電話。 觸發模組103用於當偵測模組102偵測到通話裝置1接通電 話時，觸發語音識別系統12和聲紋識別系統13開啟。通 話裝置1接通電話後，話筒11將即時採集周圍環境的聲音 。語音識別系統12開啟後，將即時從話筒11所採集的聲 音中識別用戶說話的内容。 [0016] G [0017] 偵測模組102還用於偵測語音識別系統12所識別的内容， 從而偵測用戶是否說出所述開始指令和結束指令。 用戶在通話過程中，首先選擇一個相對安靜的環境說出 開始指令和一些其他的話，直至說完結束指令。用戶也 可以在說完開始指令後直接說出結束指令。由於該開始 指令、該其他的話和該結束指令都是在相對安靜的環境 中說出，則可以認為話筒11在這段時間裏採集的聲音基 本上只有用戶的說話聲，而沒有周圍環境的雜訊。 觸發模組103還用於當偵測模組102偵測到用戶說出開始 指令時，觸發聲紋識別系統13開始提取話筒11所採集聲 100110007 表單編號A0101 第7頁/共17頁 1002016889-0 [0018] 201240423 音的聲紋特徵。 [0019] 觸發模組103還用於當偵測模組102偵測到用戶說出結束 指令時，觸發聲紋識別系統13根據所提取的聲紋特徵建 立一個背景模型。 [0020] 獲取模組104用於獲取該背景模型，並將該背景模型存入 儲存器14中。 [0021] 觸發模組103還用於當偵測模組102偵測到用戶說出結束 指令後，觸發聲紋識別系統13重新即時提取話筒11所採 集聲音的聲紋特徵，並即時根據所提取的聲紋特徵建立 聲紋模型。 [0022] 獲取模組104還用於即時獲取該聲紋模型。 [0023] 比對模組105用於即時將該聲紋模型與所述背景模型進行 比對，得出該聲紋模型中與背景模型一致和不一致的部 分，從而區分話筒11所採集聲音中用戶的說話聲和環境 的雜訊。該與背景模型一致的部分對應的聲音即認為是 用戶的說話聲，而該與背景模型不一致的部分對應的聲 音即認為是環境的雜訊。 [0024] 觸發模組103還用於控制該聲紋模型中與該背景模型一致 的部分對應的聲音透過電磁感應和模/數轉換器作用後傳 輸至對方，從而將該聲紋模型中與該背景模型不一致的 部分對應的聲音，即環境的雜訊過濾。 [0025] 觸發模組1 0 3還用於當偵測模組1 02偵測到通話裝置1掛斷 電話時，觸發語音識別系統12和聲紋識別系統1 3關閉， 100110007 表單編號A0101 第8頁/共17頁 1002016889-0 201240423 [0026] 並刪除儲存器14中的該背景模型β 參閱圖3所示，係本發明通話雜訊去除方法較佳實施方式 的流程圖。在進入步驟S1之前，設置模組1〇1先設置開始 指令和結束指令，並將該開始指令和結束指令存入儲存 器14中。 [0027] Ο [0028] [0029] Ο [0030] 步驟S1，當偵測模組1〇2偵測到通話裝置丨接通電話時， 觸發模組103觸發語音識別系統12和聲紋識別系統13開啟 。通話裝置1接通電話後，話筒丨丨將即時採集周圍環境的 聲音。語音識別系統12開啟後，將即時從話筒丨丨所採集 的聲音中識別用戶說話的内蓉。 步驟S2，當债測模組1〇2俄測到用戶說出開始指令時，觸 發模組1〇3觸發聲紋識別系統u開始提取話筒採集聲 音的聲紋特徵》 ' 步驟S3 ’當偵測模組102價測到用戶說出結束指令時，觸 發模組1G3觸發聲紋識料、統職據所提取的聲紋特徵建 立一個背景模型。, 步驟S4 ’獲取模組1()4獲取該f景模型，將該背景模型存 入儲存器14中’觸發模組m觸發聲紋識別系統13重新即 時提取話㈣所採集聲音的聲紋特徵，並即時根據所提 取的聲紋特徵建立聲紋模型。 [0031] 步驟S5 100110007 ，獲取模組1G4即時獲取該聲紋模型，比對模组 m將該聲紋模型與所述背景模型進行比對，得出該聲紋 模型中與背景棋型一致和不一致的部分該與背景模型 -致的部分對應的聲音即認為是用戶的說話聲，而該與 表單編號A0101 第9頁/共17頁 1002016889-0 201240423 背景模型不一致的部分對應的聲音即認為是環境的雜訊 〇 [0032] 步驟S6，觸發模組103控制該聲紋模型中與該背景模型一 致的部分對應的聲音透過電磁感應和模/數轉換器作用後 傳輸至對方，從而將該聲紋模型中與該背景模型不一致 的部分對應的聲音，即環境的雜訊過濾。 [0033] 步驟S7，當偵測模組102偵測到通話裝置1掛斷電話時， 觸發模組103觸發語音識別系統12和聲紋識別系統13關閉 ，並刪除儲存器14中的背景模型。 [0034] 綜上所述，本發明符合發明專利要件，爰依法提出專利 申請。惟，以上所述者僅爲本發明之較佳實施方式，本 發明之範圍並不以上述實施方式爲限，舉凡熟悉本案技 藝之人士援依本發明之精神所作之等效修飾或變化，皆 應涵蓋於以下申請專利範圍内。 【圖式簡單說明】 [0035] 圖1係本發明通話雜訊去除系統較佳實施方式的運行環境 圖。 [0036] 圖2係本發明通話雜訊去除系統較佳實施方式的功能模組 圖。 [0037] 圖3係本發明通話雜訊去除方法較佳實施方式的流程圖。 【主要元件符號說明】 [0038] 通話裝置1 [0039] 通話雜訊去除系統1 0 100110007 表單編號A0101 第10頁/共17頁 1002016889-0 201240423 [0040] 話筒 11 [0041] 語音識別系統12 [0042] 聲紋識別系統13 [0043] 儲存器14 [0044] 設置模組101 [0045] 偵測模組102 [0046] 觸發模組103 f) [0047] 獲取模組104 [0048] 比對模組105 100110007 表單編號A0101 第11頁/共17頁 1002016889-0201240423 VI. Description of the Invention: [Technical Field of the Invention] [0001] The present invention relates to a noise removal system and method, and more particularly to a communication noise removal system and method. [Prior Art] [0002] At present, the technology used to transmit the voice of the party to the other party by means of a mobile phone or a telephone is generally: the electromagnetic sound is used to convert the sound emitted by the party into an analog signal, and the analog/digital conversion is performed. The analog signal is converted into a digital signal, and the digital signal is amplified and transmitted to the other party. But during the conversation, we often encounter this situation: using a mobile phone or a telephone to talk to people in a very noisy environment, even if we yell, the other party still can't hear what we are talking about. This is because the voice of our speech is mixed with environmental noise and transmitted to the other party at the same time. Obviously, this will seriously affect the quality of the call between the two parties. SUMMARY OF THE INVENTION [0003] In view of the above, it is necessary to provide a call noise removal system and method, which can remove the noise caused by the environment during the call, and clearly transmit the user's voice to the other party. [0004] A call noise removal system is implemented in a call device, the call device further comprising a voice recognition system and a voiceprint recognition system, the call noise removal system comprising: a trigger module, configured to: when the call device is connected to the phone The triggering voice recognition system and the voiceprint recognition system are turned on; the triggering module is further configured to trigger the voiceprint recognition system to start extracting the sound collected by the microphone of the communication device when detecting that the user speaks a preset start command. The voiceprint feature is further configured to trigger the voiceprint recognition system according to the detected when the user detects that the user presets the end instruction set by the preset 100110007 form number A0101 page 4/17 pages 1002016889-0 201240423 The voiceprint feature is established - a background model, and the voiceprint recognition system is used to retrieve the voiceprint features of the sound collected by the microphone, and the voiceprint model is established according to the extracted voiceprint feature; For instantly comparing the voiceprint model with the background model, and obtaining a portion of the voiceprint model that is related to the moonscape model; Trigger module further for controlling the acoustic model pattern with the background model - sound transmission portion corresponding to the other actuator. [0006] A method for removing a call noise is applied to a call device, the call device comprising a voice recognition system and a voiceprint recognition system, the method comprising the following steps: (a) when the call device is connected to the phone, Triggering the speech recognition system and the voiceprint recognition system to be turned on; (b) when detecting that the user speaks a preset start command, 'trigger the voiceprint recognition system to start extracting the voiceprint feature of the sound collected by the microphone of the communication device; (c) When detecting that the user speaks the pre-sighing end command, the trigger voiceprint recognition system establishes a background model according to the extracted voiceprint feature; (d) triggers the voiceprint recognition system to re-acquire the sound collected by the microphone. Acoustic k feature, and instantly establish a voiceprint model according to the extracted voiceprint feature; (e) Instantly compare the voiceprint model with the background model, and obtain that the voiceprint model is consistent and inconsistent with the background model And controlling the sound corresponding to the portion of the voiceprint model that is consistent with the background model to be transmitted to the other party. Compared with the prior art, the call noise removal system and method can remove the noise caused by the environment during the call, and the user's voice is clearly transmitted to the other party. [Embodiment] 100110007 Form No. A0101 Page 5 of 17 1002016889-0 201240423 [0007] Referring to Figure 1, there is shown an operational environment diagram of a preferred embodiment of the call noise removal system of the present invention. In the present embodiment, the call noise removal system 10 operates in the communication device 1. The communication device 1 can be a mobile phone, a telephone, or the like. The communication device 1 further includes a microphone 11, a voice recognition system 12, a voiceprint recognition system 13, and a storage 14. The microphone 11 is used to collect the sound of the surrounding environment when the communication device 1 is connected to the telephone, including the voice of the user when using the communication device 1 and the noise of the environment. The voice recognition system 12 is for instantly recognizing the content spoken by the user from the sounds collected by the microphone 11. The voiceprint recognition system 13 is configured to instantly extract the voiceprint feature of the voice collected by the microphone 11 when the user speaks the set start command, such as "feed", and when the user speaks the set end command, such as "hello" A background model is established based on the extracted voiceprint features. The voiceprint recognition system 13 can establish a background model by converting the extracted voiceprint features into N-dimensional feature vectors. [0010] The voiceprint recognition system 13 is further configured to, after the user speaks the end instruction, re-acquire the voiceprint feature of the sound collected by the microphone 11 and instantly establish a voiceprint model according to the voiceprint feature. [0011] The call noise removal system 10 is configured to compare the voiceprint model with the background model in real time, and control the sound corresponding to the portion of the voiceprint model that is consistent with the background model to pass electromagnetic induction and analog-to-digital conversion. After being actuated, the device transmits to the other party, thereby filtering the sound corresponding to the portion of the voiceprint model that is inconsistent with the background model, that is, the noise of the environment. [0012] Referring to FIG. 2, it is a functional module diagram of a call noise removal system of the present invention 100110007 Form No. A0101 Page 6 of 17 1002016889-0 201240423. The call noise removal system 10 includes a setup module 101, a detection module 102, a trigger module 103, an acquisition module 104, and a comparison module 105. [0013] The setting module 101 is configured to set a start command and an end command, and store the start command and the end command in the storage 14. The start command and the end command may be the starting words commonly used when the user talks, for example, the start command may be “hello”, and the end command may be “hello” and the like. [0014] Ο [0015] detection module 102 is used to detect whether the communication device 1 is connected to the telephone, and whether the telephone is hung up when the communication device 1 is connected to the telephone. The trigger module 103 is configured to trigger the voice recognition system 12 and the voiceprint recognition system 13 to be turned on when the detection module 102 detects that the communication device 1 is turned on. After the telephone device 1 is connected to the telephone, the microphone 11 will immediately collect the sound of the surrounding environment. After the voice recognition system 12 is turned on, the content spoken by the user is instantly recognized from the sounds collected by the microphone 11. [0016] The detection module 102 is further configured to detect the content recognized by the voice recognition system 12, thereby detecting whether the user speaks the start instruction and the end instruction. During the call, the user first selects a relatively quiet environment to speak the start command and some other words until the end command is finished. The user can also directly say the end command after speaking the start command. Since the start command, the other words, and the end command are all spoken in a relatively quiet environment, it can be considered that the sound collected by the microphone 11 during this time is basically only the user's voice, and there is no surrounding environment. News. The trigger module 103 is further configured to trigger the voiceprint recognition system 13 to start extracting the sound collected by the microphone 11 when the detection module 102 detects the start instruction of the user. 100110007 Form No. A0101 Page 7 / Total 17 Pages 1002016889-0 [0018] 201240423 The voiceprint feature of the sound. [0019] The trigger module 103 is further configured to trigger the voiceprint recognition system 13 to establish a background model according to the extracted voiceprint feature when the detection module 102 detects that the user has said the end instruction. [0020] The acquisition module 104 is configured to acquire the background model and store the background model in the storage 14. [0021] The trigger module 103 is further configured to trigger the voiceprint recognition system 13 to re-acquire the voiceprint feature of the sound collected by the microphone 11 after the detection module 102 detects that the user has said the end instruction, and extract the voiceprint feature according to the sound immediately. The voiceprint feature establishes a voiceprint model. [0022] The acquisition module 104 is further configured to acquire the voiceprint model in real time. [0023] The comparison module 105 is configured to compare the voiceprint model with the background model in real time, and obtain a portion of the voiceprint model that is consistent and inconsistent with the background model, thereby distinguishing users in the sound collected by the microphone 11 The noise of the voice and the environment. The sound corresponding to the portion corresponding to the background model is considered to be the user's voice, and the sound corresponding to the portion inconsistent with the background model is considered to be ambient noise. [0024] The trigger module 103 is further configured to control the sound corresponding to the portion of the voiceprint model that is consistent with the background model to be transmitted to the other party through the electromagnetic induction and the analog-to-digital converter, thereby The corresponding part of the background model corresponds to the sound, that is, the noise filtering of the environment. [0025] The triggering module 1 0 3 is further configured to trigger the voice recognition system 12 and the voiceprint recognition system 13 to be turned off when the detecting module 102 detects that the calling device 1 hangs up the phone, 100110007 Form No. A0101 No. 8 Page 17 of 1002016889-0 201240423 [0026] And deleting the background model β in the storage 14 Referring to FIG. 3, it is a flowchart of a preferred embodiment of the method for removing the call noise of the present invention. Before proceeding to step S1, the setting module 101 sets the start command and the end command first, and stores the start command and the end command in the memory 14. [0029] [0030] Step S1, when the detecting module 1〇2 detects that the calling device is connected to the phone, the triggering module 103 triggers the voice recognition system 12 and the voiceprint recognition system. 13 is turned on. When the call device 1 is connected to the phone, the microphone will instantly capture the sound of the surrounding environment. After the speech recognition system 12 is turned on, the user's speech is recognized from the sound collected by the microphone. Step S2, when the debt testing module 1〇2 detects that the user has said the start command, the triggering module 1〇3 triggers the voiceprint recognition system u to start extracting the voiceprint feature of the microphone collecting sound “Step S3” when detecting When the module 102 detects that the user has said the end command, the trigger module 1G3 triggers the voiceprint recognition, and the voiceprint feature extracted by the official job establishes a background model. Step S4 'Acquisition module 1 () 4 obtains the f scene model, and stores the background model in the storage unit 14 'trigger module m triggers the voiceprint recognition system 13 to re-acquire the voiceprint characteristics of the collected voice (4) And immediately establish a voiceprint model based on the extracted voiceprint features. [0031] Step S5 100110007, the acquiring module 1G4 acquires the voiceprint model in real time, and the comparison module m compares the voiceprint model with the background model, and obtains that the voiceprint model is consistent with the background chess type. The inconsistent part of the sound corresponding to the part of the background model is considered to be the user's voice, and the sound corresponding to the part of the form number A0101 page 9/17 page 1002016889-0 201240423 background model is considered to be The noise of the environment [0032] Step S6, the triggering module 103 controls the sound corresponding to the part of the voiceprint model that is consistent with the background model to be transmitted to the other party through the electromagnetic induction and the analog-to-digital converter, thereby transmitting the sound The sound corresponding to the part of the pattern that is inconsistent with the background model, that is, the noise filtering of the environment. [0033] Step S7: When the detecting module 102 detects that the calling device 1 hangs up, the triggering module 103 triggers the voice recognition system 12 and the voiceprint recognition system 13 to be turned off, and deletes the background model in the storage 14. [0034] In summary, the present invention complies with the requirements of the invention patent, and submits a patent application according to law. However, the above description is only the preferred embodiment of the present invention, and the scope of the present invention is not limited to the above-described embodiments, and equivalent modifications or variations made by those skilled in the art in light of the spirit of the present invention are It should be covered by the following patent application. BRIEF DESCRIPTION OF THE DRAWINGS [0035] FIG. 1 is a diagram showing an operational environment of a preferred embodiment of a call noise removal system of the present invention. 2 is a functional block diagram of a preferred embodiment of the call noise removal system of the present invention. 3 is a flow chart of a preferred embodiment of the method for removing call noise according to the present invention. [Main component symbol description] [0038] Call device 1 [0039] Call noise removal system 1 0 100110007 Form number A0101 Page 10 / Total 17 pages 1002016889-0 201240423 [0040] Microphone 11 [0041] Speech recognition system 12 [ 0042] voiceprint recognition system 13 [0043] storage module [0044] setting module 101 [0045] detection module 102 [0046] trigger module 103 f) [0047] acquisition module 104 [0048] comparison module Group 105 100110007 Form No. A0101 Page 11 of 17 1002016889-0