Voice platform with deep voice forgery detection

2019-12-18

Artificial intelligence and machine learning are powerful tools for synthetic speech. Countless studies have shown that in state-of-the-art models, it only takes a few seconds to accurately mimic the voice rhythm and intonation of a subject. For example, Baidu's latest deep voice service can clone a voice using a 3.7-second audio sample, while a cloning implementation published in a research paper in July only takes about 5 seconds. The rapid development in this field has inspired Zohaib Ahmed, former Chief Software Engineer at Magic Leap, who recently worked at BlackBerry and Hipmunk, and Saqib Muhammad to co-found like AI, based in Ontario. The duo aims to adapt leading machine learning models used for speech synthesis into a scalable service that can clone voices from relatively small datasets. Alongside the launch of their speech synthesis product, they have also introduced a tool to detect deep voice forgery. The founders had the foresight to realize that, like any tool capable of creating convincing synthetic audio, their platform could be abused by malicious actors. According to Deeptrace, a cybersecurity startup based in Amsterdam, "deep voice forgery" is growing exponentially. In its most recent analysis in June and July, Deeptrace discovered 14,698 deepfake videos on the internet, an increase from 7,964 in December last year, representing an 84% growth in just seven months. As a result, the team released a deep voice forgery detection tool called "blyzer" a few months ago. It can predict and distinguish between real and generated voices by extracting advanced representations of speech samples. By creating a summary vector containing 256 values from a given voice file, developers can compare the similarity between two voices or infer who is speaking at any given moment. Developers have a responsibility to ensure that their creations are not exploited by malicious individuals. The two founders of the company stated, "As researchers and entrepreneurs, we have considered the benefits and risks of what we create for society. When you put your voice out on our platform, we take all necessary measures to protect the ownership of your voice and ensure it is not abused."

RECENT AI NEWS

RECENT AI TOOLS