Configuring the RTP packet size
On SIP-based deployments, audio is passed directly between the Nuance Speech Server and gateway via RTP. See Call flow for an illustration. By default, the Speech Server sends audio to the gateway every 30 milliseconds (ms). This translates to an RTP packet size of 240 bytes or samples, an optimal setting for the Speech Server. (One byte equals one sample).
Most gateways send audio to the Speech Server every 20 ms, which translates to a packet size of 160 bytes. Nuance recommends configuring your gateway to send audio to the Speech Server at the same rate. For example, for Cisco gateways, add this line to the dial peer going to Voice Platform:
codec g711ulaw bytes 240
Please refer to your vendor documentation for more information.
If it’s not practical to configure your gateway, there is no impact on the caller experience or runtime performance with mismatched incoming and outgoing packet sizes. There may be some minor degradation in whole call recordings, if these are enabled.
If necessary, you can set server.mrcp2.rsspeechsynth.rtpPacketSamples to another value on the Advanced tab for the Speech Server in the Management Station. The value is in bytes.
Algorithm
RTP packets contain G.711 audio, which uses an 8-bit sample size and an 8 kHZ sample rate. This calculation shows how 30 ms of G.711 audio equals 240 bytes in the payload of the RTP packets:
30/1000 seconds*8000 samples_per_seconds*8 bit_per_sample/8 bit_per_byte=240 bytes
