Diatheke API Reference

Only one stream at a time is allowed for a session. A previously created audio input stream must be closed before starting a new one.

The Cubic model to use for ASR. This message should always be sent before any audio data is sent.

Audio data to process. The encoding of the data should match what was specified in the Diatheke server configuration. NOTE: If the audio data is empty, the server may interpret it as the end of the stream and stop accepting further messages.

Text is the Cubic engine’s formatted transcript of pushed audio. This field will be the 1-best alternative.

The confidence score is a floating point number between 0.0 - 1.0. A score of 1.0 indicates that the ASR engine is 100% confident in the transcription.

Session ID returned from the NewSession call.

Audio data to process. The encoding of the data should match what was specified in the Diatheke server configuration. NOTE: If the audio data is empty, the server may interpret it as the end of the stream and stop accepting further messages.

The label defined in the Diatheke model. Identifies which reply in the model this message corresponds to.

The reply text as defined in the Diatheke model. This is the first message that will be received for an AudioReply. It contains the same text as the corresponding ReplyEvent in the session’s event stream.

The audio data from TTS. There can be any number of these messages for an AudioReply after the first text message and before the final end message. The encoding of the data will match what was specified in the server configuration.

Indicates that TTS has finished streaming audio for the reply. This is the last message that will be received for an AudioReply.

ID of the command that should be run. i.e. “COM01” for Command #01.

A generic map of parameters (name, value). The parameters are defined in the Diatheke model. Depending on the command, these parameters should be sent back with the CommandStatus update.

ID to keep track of the dialog state when the command is requested. This field is required in the CommandStatus message so that Diatheke can correctly update the dialog state when CommandFinished is called.

session_id should be the same as the status id returned from NewSessionResponse.

ID of the command as given in the RunCommand object.

Parameters to return to Diatheke. For example, the map might contain the entry “temperature”, which was populated with a value of “30” after the command finished. Expected parameters are defined by the Diatheke model.

Set this field with an error message if a fatal error occured while executing the command (return_status == FAILURE).

State ID from the original CommandEvent. This field is required for Diatheke to correctly update the dialog state when CommandFinished is called.

Indicates Diatheke found an actionable state in the dialog, and requests the client to perform the given command.

Users should always call CommandFinished after receiving this event so that Diatheke can update the dialog state when the command is complete.

An event indicating whether pushed text and audio was recognized by ASR and/or Diatheke.

The textual reply from Diatheke in the conversation (not to be confused with the server concepts of request and response). For example, this could be a question to solicit more information from the user, a status report, or any other reply defined by the Diatheke model. The text of this message is also provided in the AudioReply stream (if one is open).

Indicates that Diatheke is expecting user input (text or audio), which is defined by input actions in the Diatheke model.

Indicates that Diatheke has returned to the start state of the model.

Array of models available for use.

For applications that have more than one model to use for ASR/NLU. ASR grammar can vary between models, as well as sets of commands. Some applications will only have one model.

Session ID returned from the NewSession call.

User input. This could be a transcription from manually run ASR, text selected from a dropdown list, entered in a prompt, etc.

The pushed text or transcription of audio sent to Diatheke.

True if the submitted text or audio transcription was recognized by the Diatheke model as a valid intent or entity.

Text of the reply event (defined by the Diatheke model).

Label of the reply event (defined by the Diatheke model).

Session ID returned from the NewSession call.

ID of the session that will have its story changed.

ID of the story to switch to. This ID is defined by the model used to create the session.

Parameters to set in session memory before executing the specified story. Some stories in the model may make assumptions about which parameters have already been defined, so it is important to be familiar with the model requirements for any given story.

If true, the given story will not be executed until the session completes the current stories and returns back to the main story. If false, the current story in the session will be immediately interrupted to execute the specified story.

If true, once the given story has finished, Diatheke will return the session to the place in the model where it was when this request was received, and restore the parameters that were defined at that time. This is useful when the change in story represents a temporary interruption. If false, Diatheke will simply continue from the given story without trying to go back to its prior state, which is useful to make a permanent state change.

The Luna model to use for TTS (defined in the server config file).

Text to synthesize

The synthesized audio data. The data encoding will match what was specified in the server configuration.

Server that manages all of the the other components.