Try to use the offset fields to generate perfect RTP timestamps.
When this option is disabled, RTP timestamps are generated from GST_BUFFER_PTS of each payloaded buffer. The PTSes of buffers may not necessarily increment with the amount of data in each input buffer, consider e.g. the case where the buffer arrives from a network which means that the PTS is unrelated to the amount of data. Because the RTP timestamps are generated from GST_BUFFER_PTS this can result in RTP timestamps that also don't increment with the amount of data in the payloaded packet. To circumvent this it is possible to set the perfect rtptime option enabled. When this option is enabled the payloader will increment the RTP timestamps based on GST_BUFFER_OFFSET which relates to the amount of data in each packet rather than the GST_BUFFER_PTS of each buffer and therefore the RTP timestamps will more closely correlate with the amount of data in each buffer. Currently GstRTPBasePayload is limited to handling perfect RTP timestamps for audio streams.