Pindrop, a company specializing in voice security, identity verification, and fraud detection, has announced the launch of Pulse Inspect, a web-based tool designed to detect AI-generated speech in digital audio and video files. What sets Pulse Inspect apart is its high degree of accuracy, claiming to detect AI-generated speech with 99% certainty, regardless of the tool or AI model used to generate the audio. This is a significant development, as most AI vendors only release AI classifiers to detect synthetic content generated from their own tools.
One of the main reasons for the release of Pulse Inspect is the rise of audio deepfakes, which have become more prevalent on the internet due to the availability of text-based generative AI systems. These deepfakes, created using popular AI tools, can mimic the voices of celebrities, politicians, and business people, leading to the spread of misinformation and scams that can damage their reputation. In fact, Pindrop’s internal report reveals that more than 12 million American adults know someone who has had deepfakes created without their consent.
To address this growing problem, Pindrop introduced the Pulse suite of products earlier this year, initially focusing on helping enterprises detect deepfake calls to their call centers. Now, with Pulse Inspect, the company is expanding its capabilities to allow organizations to check any audio or video file for AI-generated synthetic artifacts.
The process of checking for synthetic artifacts in existing media files used to be time-consuming and required forensic examination. However, Pulse Inspect streamlines this process by analyzing the audio in a matter of seconds and providing a “deepfake score” that identifies sections containing AI-generated speech. This quick response enables organizations to take proactive measures to prevent the spread of misinformation and protect their brand credibility.
Pindrop has trained its proprietary deepfake detection model on over 350 deepfake generation tools, 20 million unique utterances, and more than 40 languages. This extensive training has resulted in a detection rate of 99% for deepfake audio, based on the company’s analysis of a dataset of around 200,000 samples. The model checks media files for synthetic artifacts every four seconds, accurately classifying deepfakes even in cases where the media contains a mix of AI-generated and genuine elements.
The detection model used by Pulse Inspect is not limited to specific AI generation tools, as Pindrop has trained it on a wide range of systems. This comprehensive approach is crucial, as there are hundreds of deepfake generator systems, including both commercial and open-source tools. By covering a broad range of generation systems, Pulse Inspect can identify synthetic audio across all platforms.
However, it is worth noting that there may be cases where the tool fails to identify deepfakes, such as files with less than two seconds of net speech or a high level of background noise. Pindrop’s CEO, Vijay Balasubramaniyan, acknowledges these limitations and assures that the company is continuously working to address them and improve detection accuracy.
Currently, Pulse Inspect is targeted at organizations such as media companies, non-profits, government agencies, celebrity management firms, legal firms, and social media networks. While the exact number of customers using the tool remains undisclosed, Balasubramaniyan has confirmed that several partners are already using the product through a volume-based annual subscription. TrueMedia.org, a free-use platform that helps critical election audiences detect deepfakes, is among the partners utilizing Pulse Inspect.
In addition to manual uploads via the web app, Pulse Inspect can also be integrated into custom forensic workflows through an API. This integration allows for bulk use cases, such as social media networks flagging and removing harmful AI-generated videos.
Looking ahead, Pindrop plans to enhance the Pulse suite by improving the explainability aspect of the tools. The company aims to develop a feature that traces back to the source of deepfake generations, providing greater transparency and accountability. Additionally, Pindrop plans to support more modalities, expanding its capabilities beyond audio and video to address emerging threats in other forms of media.
In conclusion, Pulse Inspect from Pindrop offers a powerful solution for detecting AI-generated speech in digital audio and video files. With its high degree of accuracy and comprehensive approach to deepfake detection, the tool provides organizations with the means to combat the spread of misinformation and protect their brand credibility. As deepfakes continue to evolve and pose a significant threat, tools like Pulse Inspect play a crucial role in maintaining trust and integrity in the digital landscape.