[{"data":1,"prerenderedAt":666},["ShallowReactive",2],{"site-schema":3,"article-zh-tw-ai-customer-service-voice-agent":106,"article-related-zh-tw-ai-customer-service-voice-agent":649},{"@context":4,"@graph":5},"https:\u002F\u002Fschema.org",[6,85,97],{"@type":7,"@id":8,"name":9,"alternateName":10,"description":17,"knowsAbout":18,"legalName":27,"foundingDate":28,"numberOfEmployees":29,"taxID":30,"url":31,"logo":32,"founder":33,"contactPoint":35,"address":49,"location":55,"sameAs":75},"Organization","https:\u002F\u002Fjoinx.co\u002F#organization","JoinX哲煜科技",[11,12,13,14,15,16],"JoinX","TWJOIN","哲煜科技","哲煜科技股份有限公司","TWJOIN 哲煜科技","twjoin.com","JoinX 哲煜科技（原品牌名 TWJOIN）是 2016 年在台北成立的軟體開發公司，80+ 人自有團隊，承接客製系統、APP、AI 導入與 IoT 開發，以及軟體外包與既有系統接手維運，服務 220+ 客戶、完成 520+ 專案。",[19,20,21,22,23,24,25,26],"軟體外包","客製化系統開發","APP 開發","AI 導入","AI Agent 開發","IoT APP 開發","ERP 系統整合","既有系統接手維運","JoinX Co., Ltd.","2016-04-28",80,"43774383","https:\u002F\u002Fjoinx.co","https:\u002F\u002Fjoinx.co\u002Fimages\u002Flogo-with-name.png",{"@id":34},"https:\u002F\u002Fjoinx.co\u002F#person-jack-lee",[36],{"@type":37,"contactType":38,"url":39,"telephone":40,"email":41,"areaServed":42,"availableLanguage":45},"ContactPoint","customer service","https:\u002F\u002Fjoinx.co\u002Fcontact-us","+886-2-8771-9095","service@joinx.co",[43,44],"TW","JP",[46,47,48],"zh-Hant","en","ja",{"@type":50,"streetAddress":51,"postalCode":52,"addressLocality":53,"addressRegion":54,"addressCountry":43},"PostalAddress","民生東路二段170號8樓","104","台北市中山區","台灣",[56,63,69],{"@type":57,"name":58,"address":59},"Place","JoinX 台中辦公室",{"@type":50,"streetAddress":60,"postalCode":61,"addressLocality":62,"addressRegion":54,"addressCountry":43},"台灣大道二段360號21樓C室","40453","台中市北區",{"@type":57,"name":64,"address":65},"JoinX 高雄辦公室",{"@type":50,"streetAddress":66,"postalCode":67,"addressLocality":68,"addressRegion":54,"addressCountry":43},"民族一路80號2樓之一","807","高雄市三民區",{"@type":57,"name":70,"address":71},"JoinX 東京辦公室",{"@type":50,"streetAddress":72,"postalCode":73,"addressLocality":74,"addressCountry":44},"東五反田5-22-37 Office Circle N 五反田 9樓","141-0022","東京都品川區",[76,77,78,79,80,81,82,83,84],"https:\u002F\u002Fwww.crunchbase.com\u002Forganization\u002Ftwjoin","https:\u002F\u002Fwww.facebook.com\u002FJoinX.TW","https:\u002F\u002Ftw.linkedin.com\u002Fcompany\u002F%E5%93%B2%E7%85%9C%E7%A7%91%E6%8A%80","https:\u002F\u002Fwww.youtube.com\u002Fc\u002Fjoinstuido%E5%93%B2%E7%85%9C%E7%A7%91%E6%8A%80","https:\u002F\u002Fmedium.com\u002Ftwjoin","https:\u002F\u002Fwww.104.com.tw\u002Fcompany\u002F1a2x6bjomb","https:\u002F\u002Fclutch.co\u002Fprofile\u002Fjoinx","https:\u002F\u002Fwww.goodfirms.co\u002Fcompany\u002Ftwjoin","https:\u002F\u002Ftwjoin.com",{"@type":86,"@id":34,"name":87,"alternateName":88,"jobTitle":89,"worksFor":90,"url":91,"sameAs":92},"Person","李秉哲","Jack Lee","創辦人暨執行長",{"@id":8},"https:\u002F\u002Fjack88.joinx.co\u002F",[93,94,95,96],"https:\u002F\u002Fwww.instagram.com\u002Fjack88.podcast\u002F","https:\u002F\u002Fwww.threads.net\u002F@jack88.podcast","https:\u002F\u002Fjoinx.co\u002Fjack88\u002F","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fjacklee-taiwan\u002F",{"@type":98,"@id":99,"url":31,"name":9,"alternateName":100,"publisher":101,"inLanguage":102},"WebSite","https:\u002F\u002Fjoinx.co\u002F#website",[11,13,9],{"@id":8},[103,104,105],"zh-Hant-TW","en-US","ja-JP",{"id":107,"title":108,"body":109,"category":614,"cover":615,"dateModified":616,"description":617,"extension":618,"faq":616,"hasCoverTitle":619,"locale":620,"meta":621,"navigation":622,"need":623,"path":624,"seo":625,"seoDescription":616,"seoTitle":616,"sources":626,"stem":639,"takeaways":640,"time":646,"type":647,"__hash__":648},"content\u002Fzh-tw\u002Farticle\u002Fai-customer-service-voice-agent.md","AI 語音客服怎麼導入？Voice Agent 的架構、適合場景、廠商評估與費用",{"type":110,"value":111,"toc":595},"minimark",[112,120,125,128,131,134,138,141,219,222,226,229,301,304,308,311,345,348,351,355,358,420,429,433,436,513,516,520,523,553,556,559,564,567,571,574,578,581,585,588,592],[113,114,115,119],"p",{},[116,117,118],"strong",{},"AI 語音客服能不能用，看的是三件事：來電問題夠不夠重複、電話系統接不接得上，以及轉真人的流程有沒有先設計好。"," 模型本身已經能聽懂多數中文口語，真正讓專案卡住的，通常是電話串接、延遲與責任歸屬。下面依導入的順序，說明架構、場景、評估方式與費用。",[121,122,124],"h2",{"id":123},"ai-語音客服voice-agent是什麼","AI 語音客服（Voice Agent）是什麼？",[113,126,127],{},"AI 語音客服是讓 AI 接聽電話、聽懂來電者的問題、依企業資料用語音回答的系統，答不了或需要負責的事會轉給真人。它和傳統語音選單（按 1 查詢、按 2 轉接）最大的差別，是來電者可以直接用說的。",[113,129,130],{},"傳統語音選單要求來電者照著選單走，選項一多就容易按錯、繞回原點。AI 語音客服則是先聽懂來電者想做什麼，再依知識庫回答，或帶著已經問到的資訊轉給真人，客服人員不用從頭再問一次。",[113,132,133],{},"和文字 AI 客服相比，語音多了兩個限制：回答要夠快，否則對方會以為斷線；回答要夠短，因為聽的人沒辦法回頭重看。這兩點會影響架構與驗收方式。",[121,135,137],{"id":136},"ai-語音客服的架構有哪幾種","AI 語音客服的架構有哪幾種？",[113,139,140],{},"常見有兩種架構：一種是「語音轉文字、AI 回答、文字轉語音」三段串接，另一種是語音直接進出的即時語音模型。前者每一段都能替換與檢查，後者延遲較低、語氣較自然。",[142,143,144,160],"table",{},[145,146,147],"thead",{},[148,149,150,154,157],"tr",{},[151,152,153],"th",{},"比較項目",[151,155,156],{},"三段串接",[151,158,159],{},"即時語音模型",[161,162,163,175,186,197,208],"tbody",{},[148,164,165,169,172],{},[166,167,168],"td",{},"運作方式",[166,170,171],{},"語音辨識轉成文字，AI 依知識庫回答，再合成語音",[166,173,174],{},"模型直接聽語音、直接說語音",[148,176,177,180,183],{},[166,178,179],{},"延遲",[166,181,182],{},"三段相加，較容易感覺到停頓",[166,184,185],{},"較低，對話比較順",[148,187,188,191,194],{},[166,189,190],{},"可檢查性",[166,192,193],{},"每一段都有文字紀錄，容易查是聽錯還是答錯",[166,195,196],{},"要另外存逐字稿，除錯較難",[148,198,199,202,205],{},[166,200,201],{},"替換彈性",[166,203,204],{},"辨識、模型、語音合成可以各自換廠商",[166,206,207],{},"綁定支援即時語音的模型",[148,209,210,213,216],{},[166,211,212],{},"適合",[166,214,215],{},"重視稽核、要查是哪一段出錯的場景",[166,217,218],{},"重視對話體驗、問題較單純的場景",[113,220,221],{},"兩種架構都要再接上電話系統，也就是企業原本的總機、客服中心系統或雲端電話服務。Microsoft 的 Azure AI Speech 提供語音轉文字與文字轉語音，OpenAI 的 Realtime API 提供低延遲的語音對話，選哪一種要看企業對延遲、稽核與資料存放的要求。",[121,223,225],{"id":224},"哪些來電適合交給-ai-語音客服","哪些來電適合交給 AI 語音客服？",[113,227,228],{},"適合的是重複、量大、答案明確的來電，例如查詢進度、營業時間、預約變更與常見規則說明。需要人判斷、要負法律或財務責任，或來電者情緒激動的電話，應該直接轉真人。",[142,230,231,244],{},[145,232,233],{},[148,234,235,238,241],{},[151,236,237],{},"來電類型",[151,239,240],{},"建議做法",[151,242,243],{},"原因",[161,245,246,257,268,279,290],{},[148,247,248,251,254],{},[166,249,250],{},"營業時間、地點、規則說明",[166,252,253],{},"交給 AI 回答",[166,255,256],{},"答案固定，寫在文件裡就答得出來",[148,258,259,262,265],{},[166,260,261],{},"查訂單、查維修或申請進度",[166,263,264],{},"AI 回答，先只開放查詢",[166,266,267],{},"要串接既有系統，修改類動作保留人工",[148,269,270,273,276],{},[166,271,272],{},"預約、改期、取消",[166,274,275],{},"AI 處理後由系統或人工確認",[166,277,278],{},"牽涉名額與時段，要防止重複或誤刪",[148,280,281,284,287],{},[166,282,283],{},"需要核對身分的交易",[166,285,286],{},"轉真人",[166,288,289],{},"身分驗證與責任歸屬要由人負責",[148,291,292,295,298],{},[166,293,294],{},"客訴、退款、帳務爭議",[166,296,297],{},"直接轉真人",[166,299,300],{},"需要判斷與安撫，AI 回答容易擴大問題",[113,302,303],{},"判斷的方法很簡單：把過去 3 個月的來電紀錄依問題分類，看前幾類占了多少比例，以及答案是否寫在文件或系統裡。答案找得到、問法又重複的類別，才是第一階段的候選。",[121,305,307],{"id":306},"導入-ai-語音客服前要先確認哪些事","導入 AI 語音客服前，要先確認哪些事？",[113,309,310],{},"最重要的是三件事：電話系統能不能把通話交給 AI、轉真人時能不能帶上前面的對話內容，以及錄音與逐字稿怎麼保存。這三件事沒確認，PoC 做得再好也上不了線。",[312,313,314,321,327,333,339],"ol",{},[315,316,317,320],"li",{},[116,318,319],{},"確認電話系統的串接方式。"," 問清楚現有總機或客服中心系統能不能轉接到外部服務、支援哪些通訊協定，雲端電話服務通常比傳統總機容易串接。",[315,322,323,326],{},[116,324,325],{},"設計轉真人的條件與流程。"," 哪些關鍵字、哪些情緒、問了幾次答不出來就轉真人，轉過去時客服人員要看到什麼資訊。",[315,328,329,332],{},[116,330,331],{},"整理知識庫。"," 把常見問題的答案寫成適合「聽」的版本：句子短、先講結論、數字和日期講清楚。",[315,334,335,338],{},[116,336,337],{},"決定錄音與逐字稿的處理方式。"," 依《個人資料保護法》，蒐集個人資料前要告知目的；通話開頭要說明會錄音與使用 AI，保存期限與存放位置也要先定。",[315,340,341,344],{},[116,342,343],{},"訂出驗收指標。"," 除了答對率，還要量回應延遲、聽錯率與轉真人比例，測試用真實來電錄音，而不是只用文字題目。",[113,346,347],{},"NIST 的 AI 風險管理框架把上線後的持續監控列為治理重點。語音客服上線後，要定期抽聽通話、檢查答錯與轉真人的案例，再回頭補知識庫。",[349,350],"article-cta",{},[121,352,354],{"id":353},"ai-語音客服的廠商怎麼評估","AI 語音客服的廠商怎麼評估？",[113,356,357],{},"評估時看的不是展示有多流暢，而是對方能不能接上你們的電話系統與既有資料、能不能說清楚延遲與聽錯率怎麼量，以及資料放在哪裡。展示用的題目都是挑過的，要用自己的真實來電測。",[142,359,360,370],{},[145,361,362],{},[148,363,364,367],{},[151,365,366],{},"評估項目",[151,368,369],{},"要問的問題",[161,371,372,380,388,396,404,412],{},[148,373,374,377],{},[166,375,376],{},"電話串接",[166,378,379],{},"能不能接我們現有的總機或客服中心系統？轉真人時會帶哪些資訊？",[148,381,382,385],{},[166,383,384],{},"知識與系統串接",[166,386,387],{},"回答依據哪些文件？能不能查訂單、預約等系統？查詢與修改的權限怎麼分？",[148,389,390,393],{},[166,391,392],{},"驗收方式",[166,394,395],{},"用什麼資料測？答對率、回應延遲、聽錯率與轉真人比例各怎麼量？",[148,397,398,401],{},[166,399,400],{},"資料處理",[166,402,403],{},"錄音與逐字稿存在哪裡？保存多久？會不會被拿去訓練模型？",[148,405,406,409],{},[166,407,408],{},"上線後維運",[166,410,411],{},"知識庫誰更新？答錯的通話多久檢討一次？模型更新後誰重新測試？",[148,413,414,417],{},[166,415,416],{},"交付物",[166,418,419],{},"原始碼、設定與文件歸誰？之後要換語音或模型廠商，能不能只換那一段？",[113,421,422,423,428],{},"市面上的做法大致分成兩類：一類是平台型服務，開通後設定知識庫就能用，按通話分鐘數或處理次數計費；另一類是客製整合，依企業的電話系統與既有系統開發。問題單純、不用查內部資料時，平台型比較快；要查訂單會員資料、資料不能離開指定環境時，客製整合比較合適。選擇方式和文字 AI 客服相同，可以參考 ",[424,425,427],"a",{"href":426},"\u002Farticle\u002Fai-customer-service-cost\u002F","AI 客服導入費用","裡的比較。",[121,430,432],{"id":431},"ai-語音客服的費用怎麼算","AI 語音客服的費用怎麼算？",[113,434,435],{},"客製導入的階段與文字 AI 客服相同：需求評估免費，單一場景 PoC 約 NT$30 萬至 80 萬、正式上線約 NT$80 萬至 250 萬。語音另外多出電話系統串接，以及語音辨識、語音合成或即時語音模型的用量費。",[142,437,438,451],{},[145,439,440],{},[148,441,442,445,448],{},[151,443,444],{},"費用項目",[151,446,447],{},"計費方式",[151,449,450],{},"影響金額的因素",[161,452,453,469,480,491,502],{},[148,454,455,458,466],{},[166,456,457],{},"需求評估、PoC、正式上線",[166,459,460,461,465],{},"需求評估免費，之後依階段計費，區間同 ",[424,462,464],{"href":463},"\u002Fai-customer-service\u002F","AI 客服導入","頁",[166,467,468],{},"來電類型數量、知識庫整理難度、要串接的系統",[148,470,471,474,477],{},[166,472,473],{},"電話系統串接",[166,475,476],{},"依範圍估價",[166,478,479],{},"現有總機或客服中心系統的開放程度、轉真人的設計",[148,481,482,485,488],{},[166,483,484],{},"語音與 AI 模型 API",[166,486,487],{},"依用量（通話分鐘數或處理量）",[166,489,490],{},"每月來電數、平均通話長度、選用的架構",[148,492,493,496,499],{},[166,494,495],{},"雲端電話與主機",[166,497,498],{},"依用量與規格",[166,500,501],{},"同時通話數、錄音保存量、是否需要私有部署",[148,503,504,507,510],{},[166,505,506],{},"維運",[166,508,509],{},"另訂合約",[166,511,512],{},"知識庫更新頻率、通話抽檢與報表需求",[113,514,515],{},"控制預算最有效的方法，是第一階段只開一條來電路線，例如只接「查進度」這類電話，其他電話照舊由真人接。量到答對率、延遲與轉真人比例都達標，再擴大到其他問題類型。",[121,517,519],{"id":518},"joinx-做過哪些-ai-對話與語音專案","JoinX 做過哪些 AI 對話與語音專案？",[113,521,522],{},"JoinX 公開的相關專案包括保險公司的 AI Agent 對話介面、聯經數位的 AI 語音生成平台，以及官網上的 JoinX GPT。共同做法是依既有資料與工作流程設計，AI 負責查詢與整理，需要負責的決定留給人。",[524,525,526,537,547],"ul",{},[315,527,528,531,532,536],{},[116,529,530],{},"保險公司："," 2026 年為保險業務情境設計並製作 AI Agent 的對話介面與操作流程，依第一線人員的工作順序，讓 AI 協助查詢與整理資料，最後由人確認。詳見",[424,533,535],{"href":534},"\u002Fcase-studies\u002Finsurance-ai-agent\u002F","保險公司 AI Agent 案例","。",[315,538,539,542,543,536],{},[116,540,541],{},"聯經數位："," 設計從原稿上傳、文本拆解、角色辨識、情緒分析、聲音標記到語音合成的完整流程，把電子書轉成角色鮮明的有聲內容。詳見",[424,544,546],{"href":545},"\u002Fcase-studies\u002Faicast\u002F","聯經數位案例",[315,548,549,552],{},[116,550,551],{},"JoinX GPT："," joinx.co 右下角的 AI 助手，依官網的服務、流程與文章內容回答問題，問不到的交給業務。",[113,554,555],{},"語音客服要接電話系統與既有資料，比單純的文字對話多了串接與驗收的工作。JoinX 的做法是先盤點來電類型與電話系統，再用一條來電路線做 PoC，確認數字達標才擴大。",[121,557,558],{"id":558},"常見問題",[560,561,563],"h3",{"id":562},"ai-語音客服和傳統語音選單ivr差在哪","AI 語音客服和傳統語音選單（IVR）差在哪？",[113,565,566],{},"傳統語音選單要來電者照著按鍵選項走，AI 語音客服則讓來電者直接用說的，AI 聽懂問題後回答或轉接。很多企業會保留原本的語音選單，只把其中幾個選項改由 AI 處理，逐步替換。",[560,568,570],{"id":569},"ai-語音客服聽得懂台語或口音嗎","AI 語音客服聽得懂台語或口音嗎？",[113,572,573],{},"主流語音辨識服務對國語的辨識已經相當成熟，台語與混用語言的辨識效果差異較大，要用自己的真實來電錄音測過才知道。如果來電者常說台語，PoC 時就要把台語通話放進測試集，聽錯率太高的類別先保留給真人。",[560,575,577],{"id":576},"導入-ai-語音客服客服人員會被取代嗎","導入 AI 語音客服，客服人員會被取代嗎？",[113,579,580],{},"比較常見的結果是分工改變：重複的查詢交給 AI，客服人員處理客訴、退款與需要判斷的電話。轉真人時帶上前面的對話內容，客服人員也不用從頭再問一次。",[560,582,584],{"id":583},"通話錄音交給-ai-處理要注意哪些法規","通話錄音交給 AI 處理，要注意哪些法規？",[113,586,587],{},"通話錄音與逐字稿屬於個人資料，依《個人資料保護法》，蒐集前要告知目的，使用不能超出告知範圍。實務上會在通話開頭說明錄音與使用 AI，訂好保存期限，並確認語音服務廠商不會拿資料訓練模型。金融與醫療等產業還要對照主管機關的規範。",[560,589,591],{"id":590},"ai-語音客服要多久才能上線","AI 語音客服要多久才能上線？",[113,593,594],{},"依 JoinX 的 AI 客服階段估算，需求評估約 1 至 2 次會談、單一場景 PoC 約 4 至 8 週、正式上線約 2 至 4 個月。語音專案的時程，主要看電話系統串接的難度，以及要串接幾套既有系統。",{"title":596,"searchDepth":597,"depth":597,"links":598},"",2,[599,600,601,602,603,604,605,606],{"id":123,"depth":597,"text":124},{"id":136,"depth":597,"text":137},{"id":224,"depth":597,"text":225},{"id":306,"depth":597,"text":307},{"id":353,"depth":597,"text":354},{"id":431,"depth":597,"text":432},{"id":518,"depth":597,"text":519},{"id":558,"depth":597,"text":558,"children":607},[608,610,611,612,613],{"id":562,"depth":609,"text":563},3,{"id":569,"depth":609,"text":570},{"id":576,"depth":609,"text":577},{"id":583,"depth":609,"text":584},{"id":590,"depth":609,"text":591},"技術分享","\u002Fimages\u002Fblog\u002Fai-customer-service-voice-agent.webp",null,"AI 語音客服（Voice Agent）是讓 AI 直接接聽電話、聽懂來電者的問題並用語音回答，答不了或需要負責的事再轉給真人。它適合問題重複、量大、答案寫得清楚的來電，例如查進度、營業資訊、預約變更；客訴、退款與需要身分核對的交易，仍應保留給真人。本文說明 Voice Agent 的兩種架構、導入前要確認的事、怎麼評估廠商，以及費用由哪些項目組成。","md",false,"zh-tw",{},true,"AI 客服","\u002Fzh-tw\u002Farticle\u002Fai-customer-service-voice-agent",{"title":108,"description":617},[627,630,633,636],{"name":628,"issuer":629},"個人資料保護法","法務部（全國法規資料庫）",{"name":631,"issuer":632},"Azure AI Speech 服務文件（語音轉文字、文字轉語音）","Microsoft",{"name":634,"issuer":635},"Realtime API 文件","OpenAI",{"name":637,"issuer":638},"AI Risk Management Framework（AI RMF 1.0，2023）","美國國家標準與技術研究院（NIST）","zh-tw\u002Farticle\u002Fai-customer-service-voice-agent",[641,642,643,644,645],"AI 語音客服由語音辨識、AI 回答、語音合成與電話系統串接四塊組成，難點多半在電話串接與轉真人，不在模型。","適合交給 AI 的是重複、量大、答案明確的來電；客訴、退款與需要核對身分的交易應直接轉真人。","語音比文字多兩個驗收項目：回應延遲與聽錯率，PoC 要用真實來電錄音測，不能只用文字題目測。","通話錄音與逐字稿屬於個人資料，蒐集前要告知、保存要有期限，也要決定資料放在哪裡。","費用階段與文字 AI 客服相同，另外加上語音 API 用量與電話系統串接，第一階段建議只開一條來電路線。","2026\u002F11\u002F10","blog","8RfhVTrseZj2bvgSMUtfu7MPXdRzch_NmE8nL_5bthw",[650,655,661],{"slug":651,"title":652,"time":653,"cover":654,"need":623},"ai-customer-service-cost","AI 客服導入費用：需求評估、PoC、正式上線各多少錢","2026\u002F11\u002F04","\u002Fimages\u002Fblog\u002Fai-customer-service-cost.webp",{"slug":656,"title":657,"time":658,"cover":659,"need":660},"pos-system-guide","POS 系統怎麼選？套裝、雲端 POS 與客製 POS 的差異、硬體、串接與費用（2026）","2026\u002F11\u002F12","\u002Fimages\u002Fblog\u002Fpos-system-guide.webp","POS 系統",{"slug":662,"title":663,"time":646,"cover":664,"need":665},"crm-system-selection-guide","CRM 系統怎麼選？和 ERP、會員系統的分工、導入順序與費用（2026）","\u002Fimages\u002Fblog\u002Fcrm-system-selection-guide.webp","CRM",1794558627559]