You can experiment with the impedance settings of the ATA or FXS device you've connected to the Valcom paging adapter to minimize this. If it's one-way paging, you can also set the gain towards the SIP side to a minimum value. There's no need for audio to come back to the SIP side from one-way speakers. Any echo suppression settings will be in the ATA and not in the 3120.
However, some of what your users are experiencing may not be echo at all, but latency. When RTP is converted to an analog signal, there is latency when a voice sample is collected, packetized, and sent. Another delay occurs when it is then streamed as audio. In addition, there are jitter buffers that intentionally induce delay to compensate for possible delays in getting the next packet. In a straight analog system, none of these delays exist and the voice from the loudspeakers appears to be immediate, with only the speed-of-light delay in the wire in the way.
If the person initiating the page can hear their own voice coming from the loudspeakers, it will seem to have echo due to these delays. The ear and brain don't react well to hearing one's own voice delayed, and it can be quite confusing. You can experience the same thing if you make a cell phone call to someone in the same room with you and you hear their voice both live and in your phone.
You can test to see which is happening in your system by disconnecting the loudspeakers or turning the volume of the paging amplifier all the way down. Make a page. If you still hear your own voice in the handset you have echo. If not, it's the latency due to the analog-to-RTP-to-analog conversion and not a fault in the equipment.
There are store-and-forward paging systems that mitigate this as well as prevent feedback by buffering the voice until the person initiating the page has hung up. This will eliminate the problem.
