Diatheke is Cobalt’s dialog management engine. It uses a combination of speech technologies and artificial intelligence to allow users to interact with computers and mobile devices through audio and text based dialogs.
Audio based dialog management uses multiple speech technologies, including Automatic Speech Recognition (ASR), Natural Language Understanding (NLU), and Text To Speech (TTS). It starts with audio input from a client application (human speech), which Diatheke sends to the ASR engine to get a transcription. This transcription is passed on to the NLU engine, which converts the transcription into an intent and entities. The intent and entites are then used to perform an action, as defined by the Diatheke model. One such action is for Diatheke to send a reply to the user. The reply text is sent to the TTS engine, which then synthesizes audio to send back to the client, as shown below.
Diatheke is also capable of processing dialogs without using audio input or output. In this case, text is sent to Diatheke, which forwards it directly to the NLU engine and converts the text to an intent and entities, as it did with the transcription in the audio workflow. The intent and entities are used to perform an action, such as sending a reply to the client application in the form of text, as shown below.
Diatheke uses an event stream to notify the client application when significant events happen in Diatheke. The most important of these events is the Command event, which represents a request from the server for the client application to execute a command defined in the Diatheke model. The client in turn notifies Diatheke when the command has finished, as shown below.