fix(flow): improve tail capture and final chunk ASR recovery

- Add FlowUtteranceEndCoordinator with 350ms silence drain and 150ms post-roll
- Extend FinalChunkRecovery for short/empty final chunks in chunked pipeline
- Snapshot partial at mic stop and guard final transcript in FlowSessionManager
- Unify tail drain presets (iosFlow/macMLX) and expand diagnostics

Co-authored-by: Rocky <hkgood@users.noreply.github.com>
This commit is contained in:
Cursor Agent
2026-07-26 10:13:40 +00:00
parent e6f99d2744
commit c7ee891a90
16 changed files with 485 additions and 42 deletions
@@ -115,6 +115,34 @@ final class ChunkedUtterancePipelineTests: XCTestCase {
}
XCTAssertTrue(success.text.contains("merged"))
}
func testPipelineRetriesEmptyFinalChunkWithOverlap() async {
let config = FlowUtteranceChunkConfig(
maxChunkDurationSeconds: 0.05,
overlapDurationSeconds: 10,
pauseExtensionMaxSeconds: 0,
pauseRMSThreshold: 0.02,
minFinalChunkDurationSeconds: 0.05,
sampleRate: 1_000
)
let asr = EmptyFinalRetryStubASR()
let pipeline = ChunkedUtterancePipeline(
asr: asr,
locale: Locale(identifier: "zh-Hans"),
config: config
)
let (stream, continuation) = AsyncStream<AudioBufferSnapshot>.makeStream()
continuation.yield(AudioBufferSnapshot(samples: [Float](repeating: 0.1, count: 80), sampleRate: 1_000))
continuation.yield(AudioBufferSnapshot(samples: [Float](repeating: 0.1, count: 80), sampleRate: 1_000))
continuation.finish()
let outcome = await pipeline.transcribe(stream: stream) { _ in }
guard case .success(let success) = outcome else {
return XCTFail("expected success, got \(outcome)")
}
XCTAssertTrue(success.text.contains("recovered-tail"))
}
}
private struct FailingSecondChunkASR: ASRService, @unchecked Sendable {
@@ -171,3 +199,32 @@ private struct ShortFinalMergeStubASR: ASRService, @unchecked Sendable {
return .success("short")
}
}
private struct EmptyFinalRetryStubASR: ASRService, @unchecked Sendable {
private let callIndex = OSAllocatedUnfairLock(initialState: 0)
func transcribe(
stream: AsyncStream<AudioBufferSnapshot>,
locale: Locale
) -> AsyncStream<ASREvent> {
AsyncStream { $0.finish() }
}
func cancel() {}
func transcribeChunk(samples: [Float], locale: Locale) async -> ASRChunkResult {
_ = locale
let current = callIndex.withLock { state in
let value = state
state += 1
return value
}
if current == 0 {
return .success("head")
}
if current == 1 {
return .success("")
}
return .success("recovered-tail")
}
}