Logo

AI voice cloning fuels more sophisticated scams

AI voice cloning fuels more sophisticated scams

2024-07-21

Technological advances in AI have brought convenience to consumers, but they’ve also created new possibilities for crime. Scammers are now using sophisticated AI algorithms to replicate voices. The use of AI in fraudulent activity has been reported around the world, becoming a growing concern for law enforcement.

This April, police in New Taipei’s Sanchong District received an alert from a bank. A 71-year-old man surnamed Lee was suspected of being scammed. He was about to transfer NT$380,000.

In an era of fraud and misinformation, Mr. Lee knew that he must first make sure of the recipient’s identity. He had done so, over the phone, and was about to proceed with the transfer.

Wang Zuo-ling
Sanchong precinct police chief
The man said he received a call from someone claiming to be his son. The call was conducted over the Line app. He said that the voice on the other end was his son’s, so he thought the person was his son.

Mr. Lee was convinced that the voice belonged to his son. But police suspected the use of an AI voice generator, which has increasingly been used for fraud around the world.

Host
60 Minutes
Tell me about the call that you got.

Susan Monahan
Victim
There was a young adult on the line, saying, “Grandma, I need your help.” In a frantic voice, scared, saying, “I was driving and suddenly there was a woman stopped in front of me and I hit her and they’re gonna take me to jail.”

In the U.S., a woman was scammed out of US$9,000, or NT$270,000. She believed it was going to her grandson to cover his bail.

In Japan, a scam group fabricated the voice of Takuro Morinaga, a well-known economist, to convince people to join a Line app group.

This is the voice of FTV weatherman Lin Jia-kai, saying things he never said.

By analyzing recordings of speech, AI can use deep-learning techniques to mimic any voice in seconds.

Hsu Chung-chieh
FTV IT department
For example, to fabricate Jia-kai’s voice, I uploaded two speech samples. What were these samples? Recordings of his weather reports. You’ll find that AI was really able to capture the unique qualities of his voice.

Liao Yuan-fu
National Yang Ming Chiao Tung University professor
AI can mimic a voice using just one sentence. But although the timbre will be similar, the person’s mannerisms, such as his speech habits, can’t be fully replicated. The quality of the replication depends entirely on the data it receives. The more data you feed it, the better it will do.

Given enough data, AI can mimic a voice in any language, flawlessly capturing its timbre, tone, and natural pauses.

For more Taiwan news, tune in:
Sun to Fri at 9:30 pm on Channel 152
Tue to Sat at 1 am on Channel 53

AI詐騙 科技擬聲盜臉技術日益純熟

2024-07-21

AI科技不斷突破,帶來了便利,卻也衍生了不少犯罪問題。尤其模擬聲音的技術越來越純熟,AI詐騙案其實已經開始在各國出現,也受到警方重視。

2024年4月,新北三重分局接獲銀行通報。有位71歲李姓男子,疑似遭到詐騙,要匯款38萬。

在詐騙猖獗年代,李姓男子也知道要先確認對方身分,之所以前來匯款,正是因為先通過了電話。

[[新北三重派出所所長 汪佐凌]]
“老翁說他前幾天就有接到,自稱是他兒子的這個人,打LINE給他,然後他說他有聽過,那是他兒子的聲音,他覺得那個就是他兒子。”

兒子的聲音令李姓男子深信不疑。這疑似AI偽造聲音的手法,引起警方關注,因為這正是目前國際上,新興的詐騙方式。

[[60 Minutes節目主持人]]
“請談一下您接到的電話。”

[[詐騙受害者 Susan Monahan]]
“是個年輕人,在電話中說,阿媽,幫幫我,他聽起來又急又怕,他說開車的時候有個孕婦突然出現在面前,他不小心撞到她,現在他要被抓去關。”

這位美國阿嬤付出了9千美金,相當於27萬台幣要保釋金孫。

而在日本也有集團,假造經濟學者森永的聲音,成立投資詐騙群組。

這是民視氣象主播林嘉愷的聲音,但其實他沒說過這句話。

只要收集說話者的聲音,經過模擬學習,短短幾秒 AI就能用他的聲音說話了。

[[民視資訊部經理 徐仲杰]]
“比如說嘉愷,我上傳了他兩個檔案,這兩個檔案從哪裡來呢?就是他報氣象的錄音。你會發現嘉愷的音質,它就抓到那個特色。”

[[陽明交大智能系統研究所所長 廖元甫]]
“我們現在大概一句話就可以做得到,但是做得到是說音色會像,他講話行為,譬如說他慣用語或什麼,大概不見得學得起來,完全要靠人給它資料,要餵資料給它,越多它就越像。”

再給AI多一點聲音資料,即使要它說日文也可以,音色、語氣、停頓都能維妙維肖。

更多新聞內容,請鎖定:
民視台灣台(152頻道)週日至週五晚上9:30
民視新聞台(53頻道)週二至週六凌晨1:00

Related News

Deepfake scams, the new face of online fraud

2024-07-21

Experts weigh in on how to combat AI scams

2024-07-21

Mourners tap AI to simulate late loved ones

2024-07-21