Voice authentication vendor Pindrop on Thursday added to its family of deepfake detection tools.
Pindrop Pulse Inspect, released in preview, can detect AI-generated speech in a digital audio file.
The Inspect tool follows the launch of Pindrop Pulse at the beginning of the year. It extends beyond the vendor's existing base of call centers to media organizations, nonprofits, government agencies and social networks, Pindrop said.
Pindrop Pulse is integrated with the vendor's fraud protection and authentication platform.
Pulse Inspect enables users to upload files to the Pindrop platform to determine if the files contain synthetic speech. It also provides deepfake scores.
The launch of the tool coincides with the rise of deepfake sensitivity as the U.S. general election in November draws near.
Click here to see our full
election guide.
Pindrop previously inspected a deepfake audio of presidential candidate Kamala Harris posted on X by Elon Musk and found that the audio contained partial deepfakes. It also examined the audio of the Elon Musk deepfake released July 24 and identified that the technology used to create this deepfake audio came from voice cloning vendor ElevenLabs.
Pindrop also identified a fake robocall generated with ElevenLabs that impersonated President Joe Biden before the Democratic presidential primary in January. ElevenLabs told CNN in a statement at the time that it is "dedicated to preventing the misuse of audio AI tools."
"The one thing that is pretty obvious is it is very hard now for the human ear to distinguish a synthetically generated audio versus not," said Rahul Sood, chief product officer at Pindrop, during a discussion about the dangers of deepfakes for the upcoming election. "For the human ear, it's almost impossible now to have a high level of confidence."
A necessary tool
They're fighting AI with AI.
Lisa MartinAnalyst, Forrester
Tools like Pindrop's Pulse Inspect are necessary in the age of generative AI, said Futurum Group analyst Lisa Martin.
"They're fighting AI with AI," Martin said.
The vendor's detection technology is trained on more than 350 deepfake generation tools, 20 million-plus unique utterances and over 40 languages, according to Pindrop.
"We know how powerful GenAI is. It can be used for good. It can also be weaponized, and it has been and we're seeing it," Martin said.
The new tool is also important as experts predict that deepfakes will continue to proliferate and new algorithms will make it easier for users to create audio and other deepfakes, Martin added.
Companies such as Pindrop and competitors like Resemble AI are racing to create these detection tools to meet the spread of the technology, she added.
With the launch of Pulse Inspect, Pindrop is expanding beyond the use of its technology in call centers.
"Respeecher is working with them to say, 'help us really ensure that our technology, which is synthetic voice for Hollywood and movies, help us make sure that our technology isn't misused,' so they can focus on ethical development and use of AI voice cloning technology," Martin said.
"We need these companies to be expanding their use cases, and Pulse Inspect is going to enable them to help media companies, social media networks, nonprofits, government focus on AI-generated audio," she continued.
AI-generated image of audio deepfake detector
The challenge of volume
While Pindrop is best positioned to use AI to detect deepfakes, doing it at volume might be expensive and difficult, said Forrester Research analyst Mo Allibhai.
"We're at a stage where it's expensive ... even just from an integration standpoint," Allibha said. "We have to be selective about how we deploy the technology."
When technologies like these can be deployed at the edge, like on Apple's soon-to-be-released generative AI system for iPhones, it "becomes a little bit more trivial to some of that workflow," he added.
Edge AI lightens or removes the burden of paying for cloud compute, Allibhai said.
"That will make these solutions a lot more viable long term, for sure," he said.
Pindrop Pulse Inspect has an API-driven batch-processing platform and user interface.
Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems.
Click here to see our full
