FAQs: Architect audio buffer and flush

The barge-in function allows user input, speech or DTMF, to interrupt the play of prompts. A prompt in this case can be a .wav file or text-to-speech (TTS).

The spoken utterance or DTMF tone does not have to match anything for barge-in to occur. The VXML specification has an option for requiring a grammar match, but Edge currently does not support it. Architect sets the default barge-in to true for all audio except for the Hold Music audio, which defaults to false for historical reasons.  

You can set barge-in behavior for each individual queued prompt. When the system plays the prompt, it honors the barge-in setting for each prompt. However, after barge-in triggers, it skips the currently playing prompt to be aborted, as well as any prompts that remain in the queue, regardless of their barge-in settings.

VXML’s queue-based architecture for playing audio is efficient and also key to supporting smooth concatenative audio. However, if you encounter a case that requires you to flush queued audio sooner than waiting on flow termination or the next request for user input, add a Flush Audio action to the location in the flow where you want the system to flush the audio.

At times, Architect forces audio to flush behind the scenes by injecting a VXML <field> with a 1 millisecond timeout. This action occurs in these scenarios:

• Immediately before Call Data, and Call Secure Data actions.
• Immediately before Transfer actions (all types).
• Immediately before Disconnect and Return To Agent.
• Immediately before entering the first new Task or Menu following a Set Language.
• Immediately before to playing the audio in the Play Audio On Silence.
• As part of the Detect Silence action.
  Note: The <field> timeout is configurable with the action.
• Upon entry and exit of the Hold Music action.

For more information, see Flush Audio action.

In Architect, three operations obtain user input:

• Menus
• Collect Input actions
• Bot actions, such as Amazon Lex and Google Dialogflow

Upon entry to any of these operations, the system flushes or plays any audio that is queued because of previous calls to Play Audio, Play Estimated Wait Time action, or any other element that contains audio.