Cisco introduced several AI-focused tools today at the Webex One customer conference designed to improve performance and provide a level of automation around meeting-related tasks.
The company has been running a large language model against Webex to provide things like a meeting transcript and translation, but it wanted to develop a model that was specifically designed for Webex use cases, ones where an LLM wouldn’t necessarily be appropriate. Jeetu Patel, EVP & GM for the security and collaboration business units at Cisco, says that every company is taking their own data sets and applying them to some foundational LLMs, but it wanted to create a new model designed to go further than that.
“We’re launching a real time media model, we’re calling an RMM, which sits side by side with a large language model. And it basically uses generative AI for audio and video in new and powerful ways,” Patel told TechCrunch. “This is [designed] for audio and video to add more texture and context to the text that’s there.”
He says that RMM functions involve things like noise removal, gesture recognition, speech and video enhancement, among other things.
In conjunction with the RMM, the company is also announcing an AI-powered audio codec. “We built a brand new audio codec using AI, and it’s going to actually be up to 16 times more efficient in the use of bandwidth [than prior codecs],” he said.
He says that means that instead of using 16 kilobits to send a packet, you could do it with just one kilobit, and even less than that sometimes. And not only does the new codec require a lot less bandwidth to generate higher quality audio, but when you have packet loss, the AI aspect of the codec helps recreate dropped packets to help keep the quality high.
“This AI codec will be able to make sure that you can recreate aspects of the lost packets. And so you will be able to have up to 80 or 90% packet loss at times, and still be able to go out and get crystal clear audio,” Patel said.
Another piece is what the company is calling the Webex AI Assistant, which is designed to pull together all of the AI tooling for users in a coherent way. It will provide an LLM-driven set of functions like helping craft messages, suggesting changing the tone of a message and providing meeting and message summaries in Webex, while working with RMM to provide additional functionality.
For instance, it can also be used to detect when you step away by leaving a placeholder, and then helping you fill in what you missed when you were gone with summaries, action items or a replay of the highlights of the meeting if you missed the entire thing.
Privacy often becomes a concern with AI tools like these, and Patel says the company is trying to respect customer and individual user privacy to the extent possible. “As stated in our Responsible AI Principle on Privacy ‘when AI uses personal data or makes decisions for or about a person, privacy controls must be designed into the supporting technology to assure that personal data usage is permitted, purpose-aligned, proportional and fair,’” he said.
In plain language, that means the company is saying that it puts the customer and user in control of how the data is used. The company says in the context of today’s announcements that employees will be able to turn off the “away from computer feature,” if they are worried about management surveillance. In addition, they pointed out that the company doesn’t use customer data to train its various models.
The Webex Assistant will start shipping later this year. The assistant with RMM enhancements like the “away from computer” feature will be released in the first part of 2024. The AI audio codec is scheduled to ship in the first half of 2024.