Does having a full duplex audio card help performance?

Audio Performance and Buffering

Performance is associated with the board, but also the engine. Chant components currently swap between recognition and synthesis modes because you cannot rely on the information MCI provides about the sound systems in Windows. There is no 100% sure way to know. The other thing is that some sound cards claim they are full duplex, when in reality they simulate it or attempt to.

There is a flag you can set for your session to indicate whether the Chant components should automatically handle session switching and audio sharing between recognition and synthesis.

Does the swap between modes in Chant components mean that a board that does have full duplex is faster than boards that are half duplex?

Not necessarily. There are layers of processing involved. The Chant components are doing a ChantSuspend and ChantResume under the covers. A suspend disables the active vocabularies, and pauses the recognizer. The resume, resumes the recognizer and enables the disabled vocabularies. This is to ensure the recognizer does not try to recognize the TTS audio signal.

Where the duplex performance counts is down where the engine manipulates the audio object. This is an engine layer processing issue. You will get different performance results from different engines. IBM implements its audio different from Dragon for example. Duplex audio card throughput can be faster even though your are half duplex at an application layer between recognition and synthesis. It is the audio card (driver) ability to handle all the low-level requests.

Duplex should enable it to process inbound and outbound requests more efficiently than queuing and switching.