Smart Content

Veritone has received a new US patent for its Conductor technology that performs dynamic AI model orchestration through the use of a proprietary deep neural network (DNN).

Veritone has a large funnel of pending patents and Patent No. US 11,017,780 is Veritone’s 14th US awarded patent.

Veritone’s aiWARE-based Conductor technology uses the power of its DNN to dynamically analyze data sets and apply the most optimal AI models available to that data set. The patented functionality goes beyond just efficient model selection.

The DNN uniquely looks at multiple data set features across vision, speech, text, and data sources and orchestrates a portfolio of AI models to yield the most optimal results and can efficiently combine multiple results from the models to provide the best result possible.

“A little more than a decade ago, there was only a handful of enterprise-class AI engines, but today the variety of AI engines in the market is enormous,” said Chad Steelberg, co-founder and chief executive officer of Veritone. “The aiWARE operating system with its hundreds of ready-to-deploy AI models simplifies model deployment and scaling, and the addition of Conductor further accelerates the Veritone mission to democratize AI, speeding project deployment while boosting model performance and reducing compute cost.”

The Conductor technology functions by first ingesting the data and extracting key features of that data, then reviewing the features of interest and applying previous learning to select the most accurate AI model to use on that data set from all available models in the aiWARE operating system. Model output is then evaluated and compared to an acceptance threshold. If the results are below that threshold, the learning algorithm is retrained and the process repeats until Conductor selects the most optimal model.

Veritone’s Conductor can apply optimal model selection across multiple cognitive categories within a single data set, orchestrating the models to produce the most optimal output and form a complete picture of the data. For example, telemedicine video footage can be analyzed by running face and object recognition, then transcribing and translating the voice tracks of each video. Here, Conductor applies a trained facial recognition engine that recognizes the patient’s face to locate the patient’s records. Conductor detects medical language spoken by the doctor in the video transcription and knows to use a specialized medical transcription engine. Conductor then detects the Spanish language in the transcription and calls an aiWARE Spanish-to-English translation engine to obtain an English transcript of the video conversation.

“Siloed single-engine AI solutions will not age well as new faster, cheaper, and more accurate engines come on the market. Many companies will see their ‘modern’ AI solutions become antiquated and costly to use,” said Al Brown, Veritone Chief Technology Officer. “aiWARE’s Conductor technology takes the guesswork out of selecting the right engines for the job with its dynamic learning and application of AI models to unique data sets.”

Veritone customers like Bloomberg, the San Francisco Giants and Westwood One have been leveraging the power of aiWARE for years to transform audio, video, text, and other data sources into actionable intelligence, at scale. With hundreds of engines to choose from, aiWARE enables existing AI-enabled applications to leverage new engines that become available to better match the needs of the business, without disruption. Much like the huge leaps forward experienced in computer hardware in the 1990s–with the increase in speed and decrease in the cost of memory and CPUs–applications simply ran better and faster. By leveraging Conductor’s dynamic model orchestration, AI-enabled applications will continue to learn and evolve to meet the needs of the AI-powered enterprise.