Mixing

Keep the separate tracks for transcription, and mix a copy for people to listen to:

try recording.mix(to: folder.appendingPathComponent("call.m4a"))

or for any files with the same format:

try AudioMixer.mix([micURL, systemURL], to: outputURL)

Anything else throws AudioMixer.Error.unsupportedExtension.

How it mixesCopy Anchor Link

The tracks are summed sample by sample, each scaled by 1/√n rather than 1/n. In a call usually one side talks at a time, so this keeps speech from getting quieter than in the original tracks. Peaks above 0.9 are soft clipped instead of cut off.

The tracks must share sample rate and channel count, which AudioRecorder guarantees. A shorter track counts as silence after its end.