inputtap

inputtap

Capture audio from the default microphone via AVAudioEngine + tap, resample to PCM16 mono at 24 kHz with AVAudioConverter, and print frame counts and peak amplitude every second.

Run

go run .

macOS prompts for microphone permission on first launch. Speak into the mic; expect output like:

capturing: hwRate=48000 → outRate=24000 (Ctrl-C to stop)
frames=24000 (+24000) peak=0.123
frames=48000 (+24000) peak=0.456

Higher peak values mean louder input.

Notes

• This example exercises the input side of AVAudioEngine, which works reliably under purego because tap callbacks fire on a normal dispatch queue.
• Streaming PCM to the speakers (AVAudioPlayerNode + ScheduleBuffer or AVAudioSourceNode + render block) does not currently work under purego for continuous streams under current purego bridging. Pipe PCM to an external process such as ffplay, use a small cgo shim for render callbacks, or use AudioQueueServices for realtime output.
• macOS TCC requires the binary to be code-signed for microphone access. For development, codesign the binary with a Microphone usage entitlement or wrap it with tmc/macgo.

Key APIs

• AVAudioEngine.InputNode
• AVAudioNode.InstallTapOnBusBufferSizeFormatBlock
• AVAudioConverter.ConvertToBufferErrorWithInputFromBlock

Implementation gotchas

• Call runtime.LockOSThread() at the top of main. AVFoundation's notification delivery and runloop pumping want a stable OS thread.
• The tap format must equal InputFormatForBus(bus) — Apple requires the tap and connection formats to match when both are non-nil.
• Microphone input is float32. Convert via AVAudioConverter to whatever target format you want.
• Use the input-block converter variant (ConvertToBufferErrorWithInputFromBlock), not the simpler ConvertToBufferFromBufferError — the conversion is both sample-rate AND format, and the basic variant only handles the latter.
• Use avfaudio.InstallTapOnBus, which keeps the Objective-C block alive until the returned cleanup function runs after RemoveTapOnBus.
• Set a consumed flag inside the input block so the converter sees AVAudioConverterInputStatus_NoDataNow after handing over the tap buffer once. Without it the converter spins asking for more input.