我給人工智能電話玩弄了？ 越來越搞不清電話裡是人還是機械人？Getting confused with who’s on the phone… Is that a human or a machine
第一次感到人工智能的偉大，是看到網上有工程師用人工智能創造已故球星高比拜仁的聲音(https://twitter.com/hyovictor/status/1240411281271619584?s=21)。作為超過十年高比拜仁的追隨者，幻想有一日可以利用科技將他復活。如果可以的話，同一道理，科技能夠復活其他的已過世的名人，令他們長存不朽。
One time, knowing that there are engineers created Kobe Bryant’s voice by artificial intelligence, I first feel that how powerful A.I. is. (https://twitter.com/hyovictor/status/1240411281271619584?s=21) As a huge fan of Kobe Bryant, I started to imagine if technology could bring him back to life someday. If it is possible, we can bring back other celebrities who passed away.
人工智能客服的感覺
Talking to A.I. CS
碰巧中午接了一通奇怪的來電，聽到一位女性在電話中一連串的詢問。起初因為語音平順自然，初時不以為意，直到傾談了幾分鐘之後， 當心裏的小魔怪想撩她談天的時候，才發現她不是真人客服。發現當說了題外話，心裏嘀咕 「她來來去去也是用那幾句回應哦？」， 而且最明顯是發現她的語氣及內容也完全一樣，所以她不是真人囉？
Received a strange call by chance this afternoon, the woman caller kept asking me questions. I did not realize that she was not a real human staff as her voice sounded so natural until my little monster wanted to have some more conversation with her. When I spoke something off-topic, she used the same content to reply to my question. So, she is not a real human?
智能電話機械人的聲音技術
Voice technology of Voicebot
就是這樣引發了我的好奇心，於是我去網上尋找相關資料，發現現今大部分客戶服務中心已經使用科技撥出電話，例如客戶追蹤系統 CCS、電話錄音系統VRS、追蹤和紀錄數據去協助真人客服，從而減省他們和客戶溝通時間或幫助他們整理事後資料。IVR 互動語音系統亦廣泛使用，但大多數答案都是預先錄音的吧？聲調不自然，而且好像在模仿機械人沉悶單調的發音風格，若機械人說話的聲音聽起來又會是如何呢？於是我在網上找了一找， 發現有很多人工智能創造的聲音，如果你想聽一聽人工智能創造出來的聲音，可以去Soundcloud聽一下。(https://soundcloud.com/asiabots/ai-sharon-cantonese)
This makes me curious and therefore I searched some relative contents. Most of the call centres use technology to make outbound calls, for example, CCS and VRS to track and record the data to assist the CS officers to free them from communicating with clients or sorting out records. Though IVR is commonly used. but most of them are only recording. The intonation is unnatural, it seems like they are intimidating the way of how a robot speaks. So, how does a robot actually speak? I went on searching and found that there are many voices generated by Artificial Intelligence. You could visit Sound Cloud to listen to the A.I. voice. (https://soundcloud.com/asiabots/ai-sharon-cantonese)
客服機械人其中有個技術叫 Text-to-speech (TTS Engine)，預先設計好文本就可轉換成流暢自然的聲音。 若果需要做到由A.I. 電話機械人自動按客戶問題生成回應，因為整個過程只有非常短的時間, 就需要更高層次的 Real time Neural TTS，才能夠少於一秒就生成回應。背後亦需要具備人工智能NLG自然語言生成技術，才大概能夠創造出類似於Apple Siri的回應吧。但是，很明顯Siri的答案已經預先設計好，當你重複向Siri提問相同問題時，她的答案不斷重複。期待將來NLG 技術成熟時，你每次的提問，她都可以因應不同的情況、時間甚至是心情，給予不同的答案。(可能那個時候人類會愛上與人工智能傾談？)
One of the key technology of A.I. Ambassador is Text-to-speech (TTS Engine). You could simply generate neutral speech after you pre-design a script. However, Real-time TTS is needed if the Voicebot could react instantly differently according to the question. Moreover, NLG is needed if you want to create something similar to Apple Siri. Apparently, Siri’s answers are set. When you ask similar questions to Siri, she will keep repeating her answers. Looking forward to when NLG technology is well developed that A.I. could answer you differently according to various situations, time, or even emotion. (Maybe we will enjoy chatting with A.I. at that time?)
語音機械人理解說話的根源
How could a Voicebot understand human language?
Natural Language Understanding NLU 自然語言理解就是語音機械人能「聽懂」人類千變萬化語言的秘密。NLU 並不是單單嘗試理解語句裡的關鍵字， 而是找出整句說話裏的的實體Entity和意圖intention，尋找並提供最接近 matching 的答案，所以「看起來」人工智能夠理解人類在說什麼。但事實上他並不是和人類用同樣的方法思考。令人想起1950年圖靈所提出，如何辨別一樣物件是否具有思考力的著名驗證測試。因為我們難以定義「智能」，如果一台機器與人展開對話，而人類也不能夠分辨他是人還是機械， 那麼機械人會被認為擁有「智能」。
Natural Language Understanding NLU is the secret of how Voicebot could understand human language. NLU is not only finding the keywords in the sentence, but finding the entity and intention, and then search for the answer that matches the most. So, A.I. cannot really ‘think’ like a human although it seems to be able to understand human language. It reminds me of a test, which is about identifying if an object can think, done by Turing in 1950. Since we could not identify what is ‘Artificial Intelligence’, if a machine can have a conversation with human, and the human could not identify it is a machine or not, then the machine will be defined as having intelligence.
當日這個科學界的實驗，到今天越來越接近每一位普通人； 隨著人工智能普及於每個行業，利用人工智能所撥打的電話會越來越多，很快便會有一天，到我們怎麼考驗也分不出是機械還是人的時代，那個時候除了電話客服，將會出現更多前所未有的服務或者體驗。可能到了年老的時候，在床邊伴著你喃喃細語的是人工智能。「 反正分不出來，多一個交談伴侶，在你沉悶時陪你談天說地，他見識廣博，或者會是一件有趣的事？」
With the widening usage of artificial intelligence in different industries, the calls that use A.I. to make is gaining. Sooner or later, someday we could not distinguish between human and machine, A.I. CS will provide us a spectacular service journey to us. Maybe the one who walks beside you is A.I.