Resemble AI Releases New Real-Time Deepfake Audio Detector
Generative AI speech and voice cloning startup Resemble AI has introduced a new tool for spotting synthetic audio called the Deepfake Detection Dashboard. Resemble’s Detect system aims to differentiate between real and AI-generated audio using a deep learning model developed by the startup that can process audio files and provide a prediction score judging the likelihood that it has a synthetic origin.
Resemble Detect
Resemble Detect is essentially a sophisticated AI ear that listens for the very subtle sonic artifacts inherent in any manipulated audio. Regardless of how the sound is adjusted, those clues remain, and Resemble Detect can use them to assess the likelihood that the audio is a deepfake. The Deepfake Detection Dashboard adds to Resemble’s expanding product suite for responsible and ethical use of generative AI. It is now available alongside the company’s voice cloning, speech-to-speech, and other offerings focused primarily on the enterprise market.
Resemble AI’s goal is to empower customers with an enterprise-grade solution to recognize voice-based deepfake audio across media. The company highlighted the dashboard’s scalability, accuracy, reliability, and voice isolation capabilities as key features. This is not the first experiment with deepfake detection released by Resemble AI since it launched in 2019. Resemble Detect expands upon the audio watermark feature Resemble previously introduced called the PerTh Watermarker. This tool allows Resemble AI to tag any audio produced by its software without compromising sound quality. The watermark is essentially a very soft sound that people won’t notice but that is encoded with information that a computer can decrypt and not only identify it as synthetic but trace it back to its original dataset.
“Since our inception, Resemble AI has always prioritized safety and ethics when building generative AI and this year we have released several safeguards like our neural speech PerTh Watermarker and Resemble Detect,” Resemble AI CEO Zohaib Ahmed said. “This new real-time deepfake detection dashboard is another way we can empower our customers when it comes to combating AI-based fraud effectively and ensuring legitimacy.”
Resemble first began testing the deepfake detection tool in July. At the time, Ahmed told Voicebot that it was up to 98% accurate in real-time identification of deepfake audio, though that fell to 87% when it had never encountered the voice or track. The release comes amid broader concerns about the malicious use of deepfakes and other AI synthesis technologies to perpetrate fraud or spread misinformation. Resemble AI, which focuses on developing generative voice AI applications, also recently introduced other tools to mitigate risks around artificial voices, including neural speech watermarking.
Resemble AI’s proprietary generative AI models can already train up a voice clone with just a few minutes of a person’s voice and make it speak multiple languages. Resemble has had some success with its voice clones in the entertainment industry already with the Netflix documentary, The Andy Warhol Diaries. Still, there’s a lot of demand for tools that can spot the presence of synthetic media.
Follow @voicebotaiFollow @erichschwartz
Resemble AI Creates Synthetic Audio Watermark to Tag Deepfake Speech
