79be7384dd
13 items, 555-line diff, build + 15/15 tests green.
ARCH-A3: Phase.error now carries ErrorKind (micDenied/speechDenied/asr/llm/
appGroupUnavailable/unknown) so the UI can pick icons/copy without parsing
free-form strings. Phase.ErrorKind, Phase, LLMError all Equatable.
ARCH-A4: Every TextField in APISettingsCard gets .keyboardType(.asciiCapable)
to defeat SwiftUI's iOS 18 system-keyboard hand-off that auto-suggests
Chinese/emoji and corrupts API keys / URLs / model names.
ARCH-A5 + DOC-3: PrivacyInfo.xcprivacy audited for honesty. Removed three
declared-but-unused APIs (FileTimestamp / DiskSpace / SystemBootTime) and
added ActiveKeyboards (DDA9.1) to the extension (it actually calls
advanceToNextInputMode in the tap path). Main App now declares only
UserDefaults (CA92.1). CHANGELOG updated.
ARCH-A6: Extracted PermissionManager (mic+speech permission flow, iOS 17
branching) and AppGroupPersistor (App Group load/persist) from the God
Object. KeyboardViewController drops 515 → 459 lines. KeyboardPipelineController
left in-place per risk plan — pressBegan state machine is too race-sensitive
to refactor in this pass.
RED-2: Deleted unused Theme enum (no call sites).
RED-3: Deleted unused cardStyle() alias (no call sites).
RED-7: Single source of truth for LLM timeout — LLMClient.requestTimeout +
LLMClientFactory.defaultRequestTimeout; PolishingService derives timeout from
defaultRequestTimeout+1 instead of hardcoding 15.
RED-8: ASRService.transcribe now emits .capability(onDeviceSupported:) as
first event per session; StatusBadge shows REC ⚠️ when the locale fell
back to cloud. New @Published var onDeviceSupported on State.
TEST-1: testPolishThrowsOnTransportTimeout now actually exercises
cancellation: StubURLProtocol delays response 5s, client.polish is
cancelled via Task.cancel(), test asserts the client throws .cancelled /
.transport / .decoding (was: silently passed).
TEST-2: New testPolisherSkipsNetworkWhenModeOff — PolishingService now
short-circuits when modeId == 'off' and returns trimmed input without
invoking LLMClient (proved via injected CountingLLMClient). Service was
moved to OSGKeyboardShared to be reachable from the test target.
TEST-3: KeyboardState (formerly KeyboardViewController.State) extracted
into OSGKeyboardShared so tests can @testable-import it. 5 phase/mode
tests in new KeyboardStateTests. Typealias preserves the old name.
TEST-4: New OSGKeyboardExtTests target with 6 tests covering State
initial values, phase transitions, structured-error round-trip, mode
switching, and InputMode rawValue round-trip.
167 lines
6.2 KiB
Swift
167 lines
6.2 KiB
Swift
// ASRService.swift
|
|
// OSGKeyboard · Keyboard Extension
|
|
//
|
|
// Speech-to-text abstraction over Apple's `SFSpeechRecognizer`.
|
|
// Honours a user-selected locale (auto / zh-CN / en-US / ja-JP …) so
|
|
// dictation is first-class for non-English languages.
|
|
|
|
import Foundation
|
|
import AVFoundation
|
|
import Speech
|
|
import os.lock
|
|
import OSGKeyboardShared
|
|
|
|
// MARK: - Sendable conformance
|
|
|
|
// `AVAudioPCMBuffer` and `SFSpeechRecognitionTask` are not Sendable. We
|
|
// only ever access them serially — the PCM buffer is built and consumed
|
|
// inside a single Task, and the recogniser task is cancelled but never
|
|
// shared concurrently — so an unchecked conformance is sound here.
|
|
extension AVAudioPCMBuffer: @unchecked @retroactive Sendable {}
|
|
extension SFSpeechRecognitionTask: @unchecked @retroactive Sendable {}
|
|
|
|
// MARK: - Protocol
|
|
|
|
public protocol ASRService: Sendable {
|
|
/// Start a transcription session. The returned stream emits `.partial`
|
|
/// updates and exactly one `.final` (or `.error`) before finishing.
|
|
func transcribe(
|
|
stream: AsyncStream<AudioBufferSnapshot>,
|
|
locale: Locale
|
|
) -> AsyncStream<ASREvent>
|
|
|
|
/// Cancel any in-flight recognition and tear down its tasks.
|
|
func cancel()
|
|
}
|
|
|
|
public enum ASREvent: Sendable, Equatable {
|
|
/// Emitted exactly once at the start of every `transcribe` call, so
|
|
/// the UI can flag non-on-device locales (e.g. ja-JP on devices that
|
|
/// only ship on-device ASR for en/zh). The ASR session continues
|
|
/// either way — we fall back to cloud automatically.
|
|
case capability(onDeviceSupported: Bool)
|
|
case partial(String)
|
|
case final(String)
|
|
case error(String)
|
|
}
|
|
|
|
// MARK: - Factory
|
|
|
|
public enum ASRServiceFactory {
|
|
public static func make() -> ASRService {
|
|
AppleSpeechASR()
|
|
}
|
|
}
|
|
|
|
// MARK: - Apple Speech implementation
|
|
|
|
final class AppleSpeechASR: ASRService, @unchecked Sendable {
|
|
|
|
private let lock = OSAllocatedUnfairLock()
|
|
private var recognizerTask: SFSpeechRecognitionTask?
|
|
private var feedTask: Task<Void, Never>?
|
|
|
|
func transcribe(
|
|
stream: AsyncStream<AudioBufferSnapshot>,
|
|
locale: Locale
|
|
) -> AsyncStream<ASREvent> {
|
|
AsyncStream { continuation in
|
|
let recognizer = SFSpeechRecognizer(locale: locale)
|
|
?? SFSpeechRecognizer(locale: .current)
|
|
guard let recognizer, recognizer.isAvailable else {
|
|
continuation.yield(.error("Speech recognizer unavailable for \(locale.identifier)"))
|
|
continuation.finish()
|
|
return
|
|
}
|
|
recognizer.defaultTaskHint = .dictation
|
|
|
|
let request = SFSpeechAudioBufferRecognitionRequest()
|
|
request.shouldReportPartialResults = true
|
|
request.requiresOnDeviceRecognition = recognizer.supportsOnDeviceRecognition
|
|
let onDeviceSupported = recognizer.supportsOnDeviceRecognition
|
|
if !onDeviceSupported {
|
|
#if DEBUG
|
|
print("⚠️ 设备不支持 \(locale.identifier) 端侧 ASR, 回退云端。")
|
|
#endif
|
|
}
|
|
// Tell the UI about the capability *before* any partials so
|
|
// the StatusBadge can light up the cloud-fallback indicator
|
|
// as soon as the user presses the mic.
|
|
continuation.yield(.capability(onDeviceSupported: onDeviceSupported))
|
|
|
|
let task = recognizer.recognitionTask(with: request) { result, error in
|
|
if let error {
|
|
let nsErr = error as NSError
|
|
// Codes 203 / 1110 = "no speech detected" — a normal exit.
|
|
if nsErr.code == 203 || nsErr.code == 1110 {
|
|
continuation.yield(.final(""))
|
|
} else {
|
|
continuation.yield(.error(error.localizedDescription))
|
|
}
|
|
continuation.finish()
|
|
return
|
|
}
|
|
guard let result else { return }
|
|
if result.isFinal {
|
|
continuation.yield(.final(result.bestTranscription.formattedString))
|
|
continuation.finish()
|
|
} else {
|
|
continuation.yield(.partial(result.bestTranscription.formattedString))
|
|
}
|
|
}
|
|
|
|
self.lock.withLock { self.recognizerTask = task }
|
|
|
|
// Feed audio: for each snapshot, build a 16 kHz mono Float32
|
|
// PCM buffer and immediately `request.append(pcm)`. The PCM
|
|
// buffer never leaves this task, so it doesn't need to be
|
|
// Sendable.
|
|
let feedFormat = AVAudioFormat(
|
|
commonFormat: .pcmFormatFloat32,
|
|
sampleRate: 16_000,
|
|
channels: 1,
|
|
interleaved: false
|
|
)!
|
|
self.feedTask = Task { [request] in
|
|
for await snap in stream {
|
|
if Task.isCancelled { break }
|
|
guard !snap.samples.isEmpty,
|
|
let pcm = AVAudioPCMBuffer(
|
|
pcmFormat: feedFormat,
|
|
frameCapacity: AVAudioFrameCount(snap.samples.count)
|
|
)
|
|
else { continue }
|
|
pcm.frameLength = AVAudioFrameCount(snap.samples.count)
|
|
if let dst = pcm.floatChannelData?[0] {
|
|
snap.samples.withUnsafeBufferPointer { src in
|
|
if let base = src.baseAddress {
|
|
memcpy(dst, base, snap.samples.count * MemoryLayout<Float>.size)
|
|
}
|
|
}
|
|
}
|
|
request.append(pcm)
|
|
}
|
|
if !Task.isCancelled {
|
|
request.endAudio()
|
|
}
|
|
}
|
|
|
|
continuation.onTermination = { @Sendable [weak self] _ in
|
|
self?.cancel()
|
|
}
|
|
}
|
|
}
|
|
|
|
func cancel() {
|
|
let (recTask, feedT) = lock.withLock { () -> (SFSpeechRecognitionTask?, Task<Void, Never>?) in
|
|
let r = self.recognizerTask
|
|
let f = self.feedTask
|
|
self.recognizerTask = nil
|
|
self.feedTask = nil
|
|
return (r, f)
|
|
}
|
|
recTask?.cancel()
|
|
feedT?.cancel()
|
|
}
|
|
}
|