Files
OSGKeyboard/OSGKeyboardHostSupport/Services/CloudASR/CloudASRService.swift
T
rocky be56d2210c fix(asr): keep recordings when streaming recognition fails or stalls
- Keep recording after a streaming ASR error and retry the captured PCM
  through batch ASR before reporting a failure.
- Treat an expired ASR wait as a stall so batch fallback still runs.
- Prime the managed Volcengine stream on mic touch and adopt it only while
  fresh and open.
- Auto-scroll the live transcript to the newest recognized text.
2026-10-20 20:05:08 +02:00

480 lines
19 KiB
Swift

// CloudASRService.swift
// OSGKeyboard · HostSupport
//
// Cloud-engine ASR: uploads PCM to the user's configured provider with
// personal-dictionary bias. Streaming-capable providers use one utterance
// WebSocket; others stay on chunked batch. Moonshot falls back to on-device ASR.
import Foundation
import os
#if canImport(OSGKeyboardShared)
import OSGKeyboardShared
#endif
/// A primed streaming session is opened before the utterance has an owner.
/// Route its callbacks through a mutable relay, then bind the real partial
/// consumer when the mic gesture is committed to an utterance.
private final class CloudASRPartialRelay: @unchecked Sendable {
private let lock = OSAllocatedUnfairLock<
(@Sendable (String) -> Void)?
>(initialState: nil)
func bind(_ consumer: @escaping @Sendable (String) -> Void) {
lock.withLock { $0 = consumer }
}
func publish(_ text: String) {
let consumer = lock.withLock { $0 }
consumer?(text)
}
}
/// Uploads PCM only on the user-selected cloud engine path; provider clients
/// reject missing credentials before network transmission. Personal dictionary
/// entries are sent as recognition bias. Mutable client/cancellation state is
/// lock-protected, which is the basis for `@unchecked Sendable`.
public final class CloudASRService: ASRService, @unchecked Sendable {
private struct PrimedStreamingSession {
let id: UUID
let selection: String
let localeID: String
let createdAt: ContinuousClock.Instant
let relay: CloudASRPartialRelay
let task: Task<any CloudASRStreamingSession, Error>
}
/// The gateway closes a session that receives no audio for 15 s (counted
/// from WebSocket connect). A primed session older than this is never
/// adopted, leaving headroom for the first audio frame to arrive.
public static let defaultPrimedSessionMaxAge: Duration = .seconds(10)
private let store: any ConfigurationStore
private let session: URLSession
private let localFallback: ASRService
/// Optional account-managed path. A nil value preserves existing BYOK and
/// provider-specific local-fallback selection unchanged.
private let managedClient: (any CloudASRTranscribing)?
private let managedGrants: GatewayGrantCoordinator?
private let primedSessionMaxAge: Duration
private let lock = OSAllocatedUnfairLock()
private var client: CloudASRTranscribing?
private var usesLocalFallback = false
private var boundClientSelection: String?
private var cancelled = false
private var streamingPipeline: StreamingUtterancePipeline?
private var primedStreamingSession: PrimedStreamingSession?
public init(
store: any ConfigurationStore = AppGroupStore(),
session: URLSession = .shared,
localFallback: ASRService? = nil,
managedClient: (any CloudASRTranscribing)? = nil,
managedGrants: GatewayGrantCoordinator? = nil,
primedSessionMaxAge: Duration = CloudASRService.defaultPrimedSessionMaxAge
) {
self.store = store
self.session = session
self.managedClient = managedClient
self.managedGrants = managedGrants
self.primedSessionMaxAge = primedSessionMaxAge
// `SpeechAnalyzerASR` is internal, so it can't appear in a public
// default argument value — resolve the fallback in the body instead.
self.localFallback = localFallback ?? SpeechAnalyzerASR()
}
/// Whether Flow should prefer utterance-level true streaming for the bound provider.
public var supportsUtteranceStreaming: Bool {
if store.credentialSource == .managed {
return managedClient == nil || managedClient is CloudASRStreamingCapable
}
return CloudASRModelCatalog.supportsTrueStreamingASR(for: store.asrProviderId)
}
public func resetForNewUtterance() {
lock.withLock { cancelled = false }
if usesLocalFallback {
localFallback.resetForNewUtterance()
}
}
public func warmup(locale: Locale) async {
bindClientIfNeeded()
if usesLocalFallback {
await localFallback.warmup(locale: locale)
return
}
guard let client = lock.withLock({ client }) else { return }
do {
try await client.prepare(dictionary: store.personalDictionary)
} catch {
OSGLog.asr.warning(
"cloud ASR vocabulary prepare failed: \(CloudASRLogMetadata.describe(error), privacy: .public)"
)
}
}
/// Starts the account-managed Volcengine reservation + WebSocket handshake
/// while the user's finger is still down. The utterance adopts this session
/// moments later, removing those network round trips from first-partial
/// latency without sending audio before recording actually begins.
public func primeUtteranceStreaming(locale: Locale) {
bindClientIfNeeded()
guard store.credentialSource == .managed,
let selection = lock.withLock({ boundClientSelection }),
let streamingClient = lock.withLock({
client as? any CloudASRStreamingCapable
}) else {
return
}
let localeID = locale.identifier(.bcp47)
let alreadyPrimed = lock.withLock { () -> Bool in
guard let current = primedStreamingSession else { return false }
return current.selection == selection
&& current.localeID == localeID
&& !isExpired(current)
}
guard !alreadyPrimed else { return }
cancelPrimedUtteranceStreaming()
let id = UUID()
let relay = CloudASRPartialRelay()
let dictionary = store.personalDictionary
let task = Task.detached(priority: .userInitiated) {
try await streamingClient.openStreamingSession(
locale: locale,
dictionary: dictionary,
onPartial: { relay.publish($0) }
)
}
lock.withLock {
primedStreamingSession = PrimedStreamingSession(
id: id,
selection: selection,
localeID: localeID,
createdAt: .now,
relay: relay,
task: task
)
}
// Self-expire: the keyboard's cancel command may never arrive (the
// extension can be dismissed, or a start can fail before adoption).
let maxAge = primedSessionMaxAge
Task { [weak self] in
try? await Task.sleep(for: maxAge)
guard let self, let expired = self.takePrimedSession(id: id) else { return }
FlowTrace.asr("cloud.stream.primeDiscarded", "reason=expired")
Self.close(expired)
}
FlowTrace.asr(
"cloud.stream.primeStarted",
"locale=\(localeID) selection=\(selection)"
)
}
public func cancelPrimedUtteranceStreaming() {
guard let primed = takePrimedSession() else { return }
Self.close(primed)
}
/// Removes the primed session; with an `id`, only if it is still that one.
private func takePrimedSession(id: UUID? = nil) -> PrimedStreamingSession? {
lock.withLock {
guard let current = primedStreamingSession,
id == nil || current.id == id else {
return nil
}
primedStreamingSession = nil
return current
}
}
private func isExpired(_ primed: PrimedStreamingSession) -> Bool {
ContinuousClock.now - primed.createdAt >= primedSessionMaxAge
}
/// Closes a primed session without cancelling its open task. Cancelling
/// mid-reservation can leave a reserved, never-connected gateway session
/// holding credits and a concurrency slot until its connect timeout;
/// letting the open finish and then closing releases it immediately.
private static func close(_ primed: PrimedStreamingSession) {
Task {
if let session = try? await primed.task.value {
session.cancel()
}
}
}
/// Returns the primed session only if it still matches this utterance and
/// can accept audio. Anything else is closed and the caller opens a fresh
/// session, so priming can never make an utterance fail.
private func adoptPrimedSession(
_ primed: PrimedStreamingSession?,
selection: String?,
localeID: String,
onPartial: @escaping @Sendable (String) -> Void
) async -> (any CloudASRStreamingSession)? {
guard let primed else { return nil }
guard primed.selection == selection, primed.localeID == localeID else {
FlowTrace.asr("cloud.stream.primeDiscarded", "reason=mismatch")
Self.close(primed)
return nil
}
guard !isExpired(primed) else {
FlowTrace.asr("cloud.stream.primeDiscarded", "reason=expired")
Self.close(primed)
return nil
}
primed.relay.bind(onPartial)
let session: any CloudASRStreamingSession
do {
session = try await primed.task.value
} catch {
FlowTrace.asr(
"cloud.stream.primeDiscarded",
"reason=openFailed \(CloudASRLogMetadata.describe(error))"
)
return nil
}
guard session.acceptsAudio else {
FlowTrace.asr("cloud.stream.primeDiscarded", "reason=closed")
session.cancel()
return nil
}
let ageMs = Int((ContinuousClock.now - primed.createdAt) / .milliseconds(1))
FlowTrace.asr(
"cloud.stream.primeAdopted",
"locale=\(localeID) ageMs=\(ageMs)"
)
return session
}
public func transcribeChunk(samples: [Float], locale: Locale) async -> ASRChunkResult {
guard !samples.isEmpty else { return .success("") }
if Task.isCancelled || lock.withLock({ cancelled }) { return .cancelled }
bindClientIfNeeded()
if usesLocalFallback {
return await localFallback.transcribeChunk(samples: samples, locale: locale)
}
guard let client = lock.withLock({ client }) else {
return .failure(CloudASRError.providerUnsupported.localizedDescription)
}
let startedAt = Date()
do {
let text = try await client.transcribe(
samples: samples,
sampleRate: 16_000,
locale: locale,
dictionary: store.personalDictionary
)
let trimmed = text.trimmingCharacters(in: .whitespacesAndNewlines)
FlowTrace.transcript(
"asr.cloud.chunk",
trimmed,
"engine=cloud provider=\(store.asrProviderId) samples=\(samples.count) "
+ "rms=\(FlowTrace.rms(samples)) elapsed=\(FlowTrace.seconds(since: startedAt))s"
)
return trimmed.isEmpty ? .success("") : .success(trimmed)
} catch where ProviderToolCancellation.matches(error) {
FlowTrace.asr("cloud.chunk.cancelled", "samples=\(samples.count)")
return .cancelled
} catch {
FlowTrace.warn(
"asr.cloud.chunk.failed",
"provider=\(store.asrProviderId) samples=\(samples.count) "
+ "rms=\(FlowTrace.rms(samples)) elapsed=\(FlowTrace.seconds(since: startedAt))s "
+ "\(CloudASRLogMetadata.describe(error))"
)
return .failure(error.localizedDescription)
}
}
/// Utterance-level streaming; if the session cannot start, fall back to
/// chunked batch on the same mic stream. Mid-stream failures surface as
/// errors (finalize still has PCM batch fallback).
public func transcribeUtteranceStreaming(
stream: AsyncStream<AudioBufferSnapshot>,
locale: Locale,
onPartial: @escaping @Sendable (String) -> Void
) async -> ChunkedUtterancePipelineOutcome {
bindClientIfNeeded()
// A previous utterance's abort/failure leaves `cancelled` set, which
// would make this utterance's batch fallback (`transcribeChunk`)
// return `.cancelled` for the rest of the Flow session.
lock.withLock { cancelled = false }
if usesLocalFallback {
let pipeline = ChunkedUtterancePipeline(asr: localFallback, locale: locale)
return await pipeline.transcribe(stream: stream, onPartial: onPartial)
}
guard let streamingClient = lock.withLock({ client as? CloudASRStreamingCapable }) else {
cancelPrimedUtteranceStreaming()
let pipeline = ChunkedUtterancePipeline(asr: self, locale: locale)
return await pipeline.transcribe(stream: stream, onPartial: onPartial)
}
let selection = lock.withLock { boundClientSelection }
let localeID = locale.identifier(.bcp47)
// Every utterance either adopts the primed session or closes it.
let primed = takePrimedSession()
let session: any CloudASRStreamingSession
do {
if let adopted = await adoptPrimedSession(
primed,
selection: selection,
localeID: localeID,
onPartial: onPartial
) {
session = adopted
} else {
session = try await streamingClient.openStreamingSession(
locale: locale,
dictionary: store.personalDictionary,
onPartial: onPartial
)
}
} catch where ProviderToolCancellation.matches(error) {
return .cancelled
} catch {
OSGLog.asr.warning(
"streaming ASR session open failed, using chunked batch: \(CloudASRLogMetadata.describe(error), privacy: .public)"
)
let pipeline = ChunkedUtterancePipeline(asr: self, locale: locale)
return await pipeline.transcribe(stream: stream, onPartial: onPartial)
}
let pipeline = StreamingUtterancePipeline(
client: streamingClient,
locale: locale,
dictionary: store.personalDictionary
)
lock.withLock { streamingPipeline = pipeline }
let outcome = await pipeline.transcribe(
stream: stream,
onPartial: onPartial,
preopenedSession: session
)
lock.withLock { streamingPipeline = nil }
return outcome
}
public func transcribe(
stream: AsyncStream<AudioBufferSnapshot>,
locale: Locale
) -> AsyncStream<ASREvent> {
bindClientIfNeeded()
if usesLocalFallback {
return localFallback.transcribe(stream: stream, locale: locale)
}
if supportsUtteranceStreaming, lock.withLock({ client is CloudASRStreamingCapable }) {
return AsyncStream { continuation in
continuation.yield(.capability(onDeviceSupported: false))
let task = Task {
let outcome = await self.transcribeUtteranceStreaming(
stream: stream,
locale: locale,
onPartial: { partial in
continuation.yield(.partial(partial))
}
)
switch outcome {
case .success(let success):
continuation.yield(.final(success.text))
case .failure(let message):
continuation.yield(.error(message))
case .cancelled:
break
}
continuation.finish()
}
continuation.onTermination = { @Sendable _ in
task.cancel()
self.cancel()
}
}
}
return AsyncStream { continuation in
continuation.yield(.capability(onDeviceSupported: false))
let task = Task {
var samples: [Float] = []
for await snap in stream {
if Task.isCancelled { break }
samples.append(contentsOf: snap.samples)
}
guard !Task.isCancelled, !self.lock.withLock({ self.cancelled }) else {
continuation.finish()
return
}
guard !samples.isEmpty else {
continuation.yield(.error(SharedL10n.string("error.asr.noSpeech")))
continuation.finish()
return
}
switch await self.transcribeChunk(samples: samples, locale: locale) {
case .success(let text):
if text.isEmpty {
continuation.yield(.error(SharedL10n.string("error.asr.noSpeech")))
} else {
continuation.yield(.final(text))
}
case .failure(let message):
continuation.yield(.error(message))
case .cancelled:
break
}
continuation.finish()
}
continuation.onTermination = { @Sendable _ in
task.cancel()
self.cancel()
}
}
}
public func cancel() {
lock.withLock { cancelled = true }
cancelPrimedUtteranceStreaming()
let pipeline = lock.withLock { streamingPipeline }
Task { await pipeline?.cancel() }
localFallback.cancel()
}
private func bindClientIfNeeded() {
let providerId = store.asrProviderId
let strategy = CloudASRModelCatalog.strategy(for: providerId)
let credentialSource = store.credentialSource
let selection = "\(credentialSource.rawValue):\(providerId)"
let stalePrime = lock.withLock { () -> PrimedStreamingSession? in
guard boundClientSelection != selection else { return nil }
let stalePrime = primedStreamingSession
primedStreamingSession = nil
boundClientSelection = selection
if credentialSource == .managed {
usesLocalFallback = false
client = managedClient ?? CloudASRClientFactory.make(
store: store,
session: session,
managedGrants: managedGrants
)
return stalePrime
}
usesLocalFallback = strategy == .localFallback
client = usesLocalFallback
? nil
: CloudASRClientFactory.make(store: store, session: session)
return stalePrime
}
if let stalePrime {
Self.close(stalePrime)
}
}
}