The same microphone audio read two ways: out of the encoded bitstream, and out of the samples. The bitstream column costs almost nothing and the sample column is the reference the bitstream has to agree with.
checking support…
RTCRtpScriptTransform: construct it with a Worker, assign it to
sender.transform, and read RTCEncodedAudioFrame.data —
the same Opus payload this page gets from the encoder. One rule, and it is
unforgiving: the transform is a pass-through, so every frame must be re-enqueued.
A reader that forgets silently kills the audio it was measuring.