and defined here
This is very clear and out of question.
Also, following the logs that you provided, I can tell that PCMU was used in the communication between CVP and UniMRCP server. In fact, there were two SIP/MRCPv2 sessions established: one for speech synthesis and the other for speech recognition. The latter was not even used.
Here are some logs regarding RTP streams.
2018-09-11 16:22:31:999214 [INFO] Media Path 0x7f00c4001a58 Source->[LPCM/8000/1]->Bridge->[LPCM/8000/1]->Encoder->[PCMU/8000/1]->Sink
2018-09-11 16:22:32:019189 [INFO] Open RTP Receiver
10.3.74.80:5002 <-
10.3.74.99:16764 playout [50 ms] bounds [0 - 600 ms] adaptive [1] skew detection [1]
2018-09-11 16:22:32:019197 [INFO] Media Path 0x7f00c4003d78 Source->[PCMU/8000/1]->Decoder->[LPCM/8000/1]->Bridge->[LPCM/8000/1]->Sink
You can see that no RTP packets were received, which is expected, as no RECOGNIZE was reached the server. In the meantime, the server sent 91 RTP packets, which I strongly believe contained the intended audio data.
BTW, you can make a network capture and not only follow the RTP stream but also listen to the synthesize audio.