Properties in MRCP/VXML/API

One difficult thing about developing applications using standards such as MRCP and VXML is that the names of similar parameters are often slightly different between the standards. This is particularly important when tuning voice activity detection parameters.

The following table presents a comparison of VAD properties in MRCP and their equivalent name in VXML and the LumenVox API (used with LV_SRE_StreamSetParameter and LVSpeechPort::StreamSetParameter).

In some cases the parameters do not have direct matches, so the exact mappings may differ depending on how a platform interprets the meaning of one specification or the other. Thus you should consider this table an informational guide rather than as an official interpretation of the standards.

MRCP	VXML	API
sensitivity-level	sensitivity	STREAM_PARM_VAD_VOLUME_SENSITIVITY
speech-complete-timeout	completetimeout	STREAM_PARM_VAD_EOS_DELAY
speech-incomplete-timeout	incompletetimeout	Not supported by LumenVox
com.lumenvox.end-of-speech-timeout	maxspeechtimeout	STREAM_PARM_END_OF_SPEECH_TIMEOUT
no-input-timeout	timeout	STREAM_PARM_BARGE_IN_TIMEOUT
confidence-threshold	confidencelevel	N/A (handled by application)
speed-vs-accuracy	speedvsaccuracy	Not supported by LumenVox.
n-best-list-length	maxnbest	PROP_EX_MAX_NBEST_RETURNED
recognition-timeout	N/A	N/A
waveform-url	N/A	N/A
dtmf-interdigit-timeout	interdigittimeout	N/A
dtmf-term-timeout	termtimeout	N/A
dtmf-term-char	termchar	N/A

Note that the end of speech timeout (maxspeechtimeout in VXML) is not supported by default in MRCP; it is a LumenVox vendor-specific parameter.

Also note that the sensitivity-level (sensitivity in VXML) works different with MRCP/VXML than in our API. In our API, setting this value higher makes it less sensitive to barge-in. MRCP and VXML are the opposite.