Cognitive engines are categorized into engine classes based on the type of data that they analyze.
Class
Description
Audio
The input to engines in the audio class is an audio or video file or stream. Audio engines can recognize a specific audio segment, such as an advertisement, identify sounds like a baby's cry, or detect the presence of audio or music in an audio or video file.
Biometrics
The input to engines in the biometrics class can be an image, speech, or other audio or video file or stream. By definition, the biometrics class covers cognitive analysis related to data points from the human body. Biometrics engines can detect or recognize faces, identify face attributes to estimate a person's age or ethnicity, or verify a person based on their unique iris.
Data
The inputs to engines in the data class can be structured or unstructured data. Examples include geolocation information, historical weather data, network usage data, billing records, or signals from Internet of Things (IoT) devices. Data engines can detect outliers or anomalies in data, correlate two or more data sets, identify the geographic location of a person, predict future trajectories based on historical trends, optimize presentation of content or advertising, or suggest a decision path for example.
Speech
The input to engines is an audio or video of human speech in the form of a file or a stream. Speech engines can make predictions about what was said by one or more speakers, identify those speakers, identify the language spoken, or detect vocal emotion.
Text
The input to engines in the text class can be structured or unstructured text. In some cases the text input is structured in .aion format, such as the output from a transcription (speech) engine is fed into a translation (text) engine. Examples of text engine capabilities include translating text from one language to another, summarizing it, detecting profane language, and extracting sentiment or entities. Text analytics is often used as an umbrella term for combinations of capabilities within the text class.
Transformation
The inputs to engines in the transformation class are varied and can include images, audio, video, and data. Transformation engines convert or manipulate the original input in some way, often outputting a derivative file based on the source. Examples include redacting faces or credit card numbers, summarizing or condensing a video, converting file format types, or removing extraneous noise from audio.
Utility
Engines in the utility class are internal engines or utility engines for preparing or processing cognition data.
Vision
The input to engines in the vision class is an image or video, file, or stream. Examples include recognizing objects within an image, classifying the entire image, reading license plates, or classifying actions or gestures. Most capabilities associated with the field of computer vision are included in the vision engine class as defined by Veritone, although face recognition and associated capabilities have been classified under the biometrics class.
