Check out the on-demand sessions from the Low-Code/No-Code Summit to learn how to successfully innovate and achieve efficiency by upskilling and scaling citizen developers. Watch now.
Deepgram today added three additional capabilities to an automated speech recognition (ASR) platform that is invoked via a set of application programming interfaces (APIs).
The company is adding a conversational AI option that reduces response lag to less than 300 milliseconds. That level of interaction will make it possible for organizations to deploy applications that enable humans to have highly interactive conversations with a machine in use cases involving, for example, virtual agents handling tasks related to billing, support, sales, compliance, and identity verification.
In the absence of this capability, interactions between end users and machines become disjointed to the point that the application is perceived to be broken, Deepgram CEO Scott Stephenson said.
That capability will also make it simpler to tune a base model for conversational AI to address a wider range of use cases in domains that employ unique jargon and nomenclature. The word error rate has been reduced by up to 50%, while noise and filtering crosstalk is reduced to help make key terms and phrases more easily understood.
Intelligent Security Summit
Learn the critical role of AI & ML in cybersecurity and industry specific case studies on December 8. Register for your free pass today.
In addition to the conversational AI capability, Deepgram is adding a module to enable sales and support interactions to generate offers and alerts in real time. This will also enable sales teams to improve coaching and analytics.
Finally, Deepgram is adding support for a streaming capability to process data and create transcripts in near real time.
At the core of the Deepgram platform is a set of deep learning algorithms that learn end users’ phonetic patterns. That makes it possible to continuously train a model in a way that improves accuracy over time. The platform itself can be deployed in the cloud or an on-premises IT environment.
Stephenson said that with a less than 300 millisecond lag time and more than 90% accuracy rate for transcription, conversational AI is now entering a new phase. Rather than being used primarily by enterprises as a tool to reduce customer support costs, conversational AI will be employed more to proactively engage customers, thanks in part to cloud services being able to scale more affordably, Stephenson said.
Human versus machine
Instead of viewing conversational AI as a replacement for an interactive voice response (IVR) system, in many cases conversational AI platforms will play a larger role in sales as they continue to evolve, Stephenson said.
The level of comfort with conversational AI naturally varies by individual. Some people will always prefer to engage with another human. However, just as many people now prefer to have an issue addressed without ever engaging with a sales or support team member. And as conversational AI continues to advance, it may not be so easy to determine when customers are talking to a machine versus an actual person.
Those machines are also likely to be less biased toward outcomes, Stephenson said. Humans on a support team are likely to be incentivized to either end a support call as quickly as possible or convince a customer to buy an additional service they may not need. Of course, AI models can be biased as well, but conversational AI models may eventually prove to be more consistent, Stephenson said. “A machine is more interested in objective truth,” he asserted.
This next phase of conversational AI isn’t going to arrive overnight, but as these platforms continue to evolve, the day when almost everyone will regularly engage with one or more conversational AI platforms may not be all that far off.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.