Files
Grant Gilliam 863136aeec Phases 2-6: detection, visual timeline, backend hand-off, voiceprints
Phase 2 (call detection): CallDetector using CoreAudio per-process mic
attribution (anarlog technique) — robust start+stop for Zoom/Teams/Signal/Meet,
ignoring our own recording; auto-record toggle. Built; pending live multi-app
confirmation by the user.

Phase 3 (visual timeline foundation): AppAdapter protocol + SpeakerObservation,
TimelineBuilder (hysteresis/overlap/self-merge/aliases), VisualTimeline (schema
1.1), TextRecognizer (Vision OCR), FrameSampler + GridCallAnalyzer (name OCR +
saturated-highlight active-speaker attribution), SignalAdapter, VisualObserver
(window capture; frames released, never saved; minimized->visual_gap, idle != gap).
Synthetic-frame tested; adapter geometry pending real Signal fixtures + live
VisualObserver validation.

Phase 5 (backend hand-off): SparkControlClient (multipart label-merge, sequential,
TLS-skip, 503 Retry-After/413), SessionPackager (chunk plan + WAV slice + timeline
slice/rebase), TranscriptAssembler + SpeakersFile, TranscriptPipeline. Validated
END-TO-END against the live backend (chunk -> label-merge -> speakers.json).

Phase 6 (voiceprints): VoiceprintStore (known_voiceprints, persist named
fingerprints, skip Unknown). Wired: 'Send to backend' button + transcript status,
auto-send toggle (default off) + self-name setting.

All adversarial-review findings fixed. App + XCTest suite build; tests pass.
2026-06-06 00:15:49 -05:00

71 lines
2.9 KiB
Swift

import AVFoundation
/// Converts arbitrary input PCM buffers to **16 kHz mono Float32**, maintaining
/// resampler state across calls. Reuse one instance per source stream so the
/// internal sample-rate converter stays continuous across buffers.
///
/// Not thread-safe: use one instance from a single thread. Both the mic and
/// system instances are driven exclusively from `AudioRecorder.ioQueue` (one per
/// source stream), kept continuous across buffers.
final class Resampler {
/// The canonical Phase-1 audio format: 16 kHz, mono, Float32, deinterleaved.
static let targetFormat = AVAudioFormat(
commonFormat: .pcmFormatFloat32,
sampleRate: 16_000,
channels: 1,
interleaved: false)!
private var converter: AVAudioConverter?
private var sourceFormat: AVAudioFormat?
private var ended = false
/// 16 kHz mono buffer for `input`, or nil if conversion produced nothing.
func resample(_ input: AVAudioPCMBuffer) -> AVAudioPCMBuffer? {
guard !ended, input.frameLength > 0 else { return nil }
if converter == nil || sourceFormat != input.format {
let c = AVAudioConverter(from: input.format, to: Self.targetFormat)
// Highest-quality sample-rate conversion: best anti-aliasing on the
// 48k16k downsample, which avoids harsh artifacts on loud/bright speech.
c?.sampleRateConverterQuality = .max
c?.sampleRateConverterAlgorithm = AVSampleRateConverterAlgorithm_Mastering
converter = c
sourceFormat = input.format
}
guard let converter else { return nil }
let ratio = Self.targetFormat.sampleRate / input.format.sampleRate
let capacity = AVAudioFrameCount((Double(input.frameLength) * ratio).rounded(.up)) + 64
guard let output = AVAudioPCMBuffer(pcmFormat: Self.targetFormat, frameCapacity: capacity) else {
return nil
}
var consumed = false
var error: NSError?
let status = converter.convert(to: output, error: &error) { _, inputStatus in
if consumed { inputStatus.pointee = .noDataNow; return nil }
consumed = true
inputStatus.pointee = .haveData
return input
}
if status == .error || output.frameLength == 0 { return nil }
return output
}
/// Flush the converter's internal tail at end of stream (call once on stop).
func drain() -> AVAudioPCMBuffer? {
guard !ended, let converter else { ended = true; return nil }
ended = true
guard let output = AVAudioPCMBuffer(pcmFormat: Self.targetFormat, frameCapacity: 8192) else {
return nil
}
var error: NSError?
let status = converter.convert(to: output, error: &error) { _, inputStatus in
inputStatus.pointee = .endOfStream
return nil
}
if status == .error || output.frameLength == 0 { return nil }
return output
}
}