Response time

The response time is related to the integrated system and indicates the time needed to recognise a word/sentence and issue the corresponding action. It is computed as the time between the end of the spoken word and the beginning of the action. In general the speech recognition process consists of two separate modules: a real-time  one concerns the signal acquisition and analysis and a second module - non-real time - for the speech comparison with references. Usually two implementations are used:

1. The first approach starts the recognition process (signal analysis, parameter computation, comparisons, etc.) as soon as the speech detector indicates that the speaker is uttering something. It stops when the end-point detection declares that the end of a word is detected and a valid word is recognised. The speech recognition runs synchronously with speech input acquisition. The response time  depends on the time needed for word detection and the comparisons.
2. The second approach consists of initiating the speech recognition when the system declares that it detects both word end-points (beginning and ending times).

The application developer has to know what his system should tolerate as recognition response time  and should take into consideration the different processes that may run on the same machine.

The application developer has to know to manage his CPU load when using a multi-channel system   and should require a uniform and coherent response time  on each channel . The technology provider should guarantee maximum response time  in the worst conditions. This of course depends on the application: a response time  may be acceptable for single isolated words  and may not be acceptable for recognition of a sequence of words (connected digits or words).