audiobuffer

type Config struct {
	// SampleRate overrides the recording sample rate in Hz; 0 uses the pipeline
	// output rate from the StartFrame.
	SampleRate int
	// NumChannels is 1 for a mono mix of user and bot, or 2 for stereo with the
	// user on the left and the bot on the right; 0 uses 1.
	NumChannels int
	// BufferSize is the byte threshold at which OnAudioData and OnTrackAudioData
	// fire with the audio accumulated so far; 0 emits only when recording stops.
	BufferSize int
	// AutoStart begins recording as soon as the pipeline starts.
	AutoStart bool

	// OnRecordingStarted is called when recording transitions to active.
	OnRecordingStarted func()
	// OnRecordingStopped is called after recording stops and the final audio has
	// been delivered.
	OnRecordingStopped func()
	// OnAudioData receives the merged recording (mono mix or interleaved stereo)
	// at each BufferSize boundary and once more on stop.
	OnAudioData func(audio []byte, sampleRate, numChannels int)
	// OnTrackAudioData receives the separate, time-aligned user and bot tracks at
	// each BufferSize boundary and once more on stop.
	OnTrackAudioData func(user, bot []byte, sampleRate, numChannels int)

	// EnableTurnAudio delivers each turn's audio on its own, as that turn ends,
	// through OnUserTurnAudio and OnBotTurnAudio. It is off by default: the
	// session tracks are what a recording wants, and a turn's audio is for
	// something that works turn by turn, scoring one utterance or handing it to a
	// classifier. The session tracks cannot be cut into turns afterwards, since
	// nothing in them marks where a turn began.
	//
	// It needs the worker to be tracking turns, which it does by default, since
	// the turn tracker is what says where a turn ended.
	EnableTurnAudio bool
	// OnUserTurnAudio receives everything the user said during one turn, mono,
	// once that turn ends.
	OnUserTurnAudio func(d TurnAudioData)
	// OnBotTurnAudio receives everything the bot said during one turn, mono, once
	// that turn ends.
	OnBotTurnAudio func(d TurnAudioData)

	// OnUserTurnAudioData receives one run of the user's speech, mono, each time
	// they stop speaking. A turn can hold several runs, so it fires more than
	// once per turn and says nothing about which turn the audio belongs to; use
	// OnUserTurnAudio for that.
	OnUserTurnAudioData func(audio []byte, sampleRate, numChannels int)
	// OnBotTurnAudioData receives one run of the bot's speech, mono, each time it
	// stops speaking. See OnUserTurnAudioData.
	OnBotTurnAudioData func(audio []byte, sampleRate, numChannels int)
}

type Processor struct {
	*processor.Base
	// contains filtered or unexported fields
}

type TurnAudioData struct {
	// TurnNumber is the turn this audio belongs to.
	TurnNumber int
	// Audio is everything the speaker said during the turn, as raw PCM, with the
	// pauses between their runs of speech left out.
	Audio []byte
	// SampleRate is the rate of the audio, in Hz.
	SampleRate int
	// NumChannels is how many channels the audio has.
	NumChannels int
}