The key part of the current timestamp correction algorithm is:

        // calc the right diff by audio sample rate
        if (msg->is_audio() && sample_rate > 0) {
            delta = (int64_t)(delta * 1000.0 / sample_rate);
        } else if (msg->is_video() && frame_rate > 0) {
            delta = (int64_t)(delta * 1.0 / frame_rate);
        } else {
            delta = DEFAULT_FRAME_TIME_MS;
        }

That is, it will be calculated based on the fps and sample rate. But in reality, this is not accurate. In other words, the server cannot guarantee that a problematic stream has the correct timestamp. On the contrary, if the timestamp of the stream is incorrect, an obvious correction method should be used to indicate that the stream has an anomaly. Therefore, using the default 10ms is sufficient, and when the timestamp jumps, it is corrected to 10 milliseconds.

TRANS_BY_GPT3