Waveform logging
Voice Platform allows you to record the audio from conversations as waveform logs—audio recordings of caller utterances. By default, the system records all audio that is sent to the recognizer after endpointing (that is, with leading and trailing silences removed). It does not record the raw utterances before endpointing.
Voice Platform offers several parameters that affect how endpointed waveforms are recorded:
- swirec_waveform_begin_silence sets the duration of leading silence in a recording
- swirec_waveform_end_silence sets the duration of trailing silence
- swirec_waveform_interword_max sets the maximum silence recorded between words
- swirec_waveform_speech_thresh sets the sound level that defines background noise, such that sounds beneath this threshold are treated as silence rather than speech.
It is recommended that you use the default values for these parameters, since any changes to them can make your recordings more difficult to understand. However, if necessary you can use any of them in a <property> element by adding a “swirec.” prefix to the property name, as shown below:
<property name="swirec.swirec_waveform_end_silence" value="1000"/>
