Files
OSGKeyboard/OSGKeyboardHostSupport/Services/CloudASR/CloudASRClients.swift
T
rockyandClaude Opus 5 0023741455 feat(app): add crash visibility, memory relief, and clipboard reply work
Squashes the working tree that had accumulated on cloud-agent. Two
strands of work, kept in one commit at the maintainer's request.

Crash and stability (this session):

- Subscribe to MetricKit so crash, hang, CPU, and disk-write diagnostics
  are recorded on device. Keyboard-extension memory kills are the most
  likely way a user loses the keyboard mid-sentence and iOS does not
  classify them as crashes, so they never reach App Store Connect.
- Upload those reports to POST /v1/analytics/diagnostics from internal
  builds only (local Debug and TestFlight), reusing the analytics
  installation identifier and skipping entirely when analytics are off.
  App Store builds keep everything on device.
- Shed the Chinese engine, English lexicon, and in-flight pipeline work
  at the 40 MiB and 48 MiB budget thresholds instead of waiting for a
  system memory warning that jetsam often preempts.
- Install audio taps with the input node's live format through one
  shared adaptive resampler, so an audio-route change mid-recording no
  longer raises an uncatchable format-mismatch exception. Covers the
  dictation preview and the Mac recorder, which still used the pattern
  the Flow capture path had already moved away from.
- Harden the librime bridge: never pass a nil string across a nonnull
  boundary when a user dictionary holds malformed UTF-8, null-check the
  engine API vector, and distrust a candidate count without its array.
- Guard the expanded candidate grid against a list that shrinks while
  the collection view is still laying out against the previous count.
- Resolve localized text without the App Group so the screen shown when
  that container is missing renders its message instead of crashing on
  launch, which is the one situation it exists to handle.
- Gate CI on the App Group and Keychain wiring that the keyboard
  extension hard-depends on, with a self-test proving the check fails.

Clipboard, reply, and semantics (pre-existing working tree):

- Unify the reply center, refresh keyboard AI features, and add native
  event and reminder exporters.
- Expand clipboard semantic routing with authored corpora and a
  product-blind holdout, plus the scripts that build and validate them.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-18 17:31:29 -07:00

522 lines
19 KiB
Swift

// CloudASRClients.swift
// OSGKeyboard · HostSupport
//
// Provider-specific cloud ASR backends with personal-dictionary bias.
import Foundation
#if canImport(OSGKeyboardShared)
import OSGKeyboardShared
#endif
public protocol CloudASRTranscribing: Sendable {
func prepare(dictionary: PersonalDictionary) async throws
func transcribe(
samples: [Float],
sampleRate: Int,
locale: Locale,
dictionary: PersonalDictionary
) async throws -> String
/// Settings "validate connection" probe. Verifies transport + auth only.
func probeConnection() async throws
}
extension CloudASRTranscribing {
/// Default probe: transcribe ~1 s of near-silence. An empty transcript
/// counts as success — HTTP/streaming providers only need to prove that
/// transport + auth work. Providers that reject silent/short audio or
/// close after an empty final (DashScope realtime, Volcengine SAUC)
/// override with a handshake-only probe.
public func probeConnection() async throws {
do {
_ = try await transcribe(
samples: [Float](repeating: 0.01, count: 16_000),
sampleRate: 16_000,
locale: Locale(identifier: "zh-CN"),
dictionary: .empty
)
} catch CloudASRError.emptyTranscript {
return
}
}
}
public enum CloudASRClientFactory {
public static func make(
store: any ConfigurationStore,
session: URLSession = .shared,
managedGrants: GatewayGrantCoordinator? = nil
) -> CloudASRTranscribing {
if store.credentialSource == .managed {
return ManagedVolcengineASRClient(
grants: managedGrants ?? GatewayGrantCoordinator(),
session: session
)
}
let providerId = store.asrProviderId
let strategy = CloudASRModelCatalog.strategy(for: providerId)
let asrModel = store.asrModel.isEmpty
? CloudASRModelCatalog.defaultModel(for: providerId)
: store.asrModel
switch strategy {
case .zhipuHotwords:
return ZhipuCloudASRClient(
apiKey: store.asrApiKey,
model: asrModel,
session: session
)
case .bailianStreaming:
return BailianRealtimeASRClient(
apiKey: store.asrApiKey,
endpoint: store.asrBaseURL,
model: asrModel,
vocabularyID: nil,
session: session
)
case .prompt:
return PromptCloudASRClient(
providerId: providerId,
baseURL: store.asrBaseURL,
apiKey: store.asrApiKey,
model: asrModel,
session: session
)
case .openRouterJson:
return PromptCloudASRClient(
providerId: providerId,
baseURL: store.asrBaseURL,
apiKey: store.asrApiKey,
model: asrModel,
session: session,
requestFormat: .openRouterJson
)
case .volcengineStreaming:
return VolcengineCloudASRClient(
apiKey: store.asrApiKey,
endpoint: store.asrBaseURL,
resourceID: asrModel,
session: session
)
case .openaiRealtimeStreaming:
return OpenAIRealtimeASRClient(
apiKey: store.asrApiKey,
endpoint: store.asrBaseURL,
model: asrModel,
batchBaseURL: LLMProvider.provider(id: "openai").defaultBaseURL,
session: session
)
case .localFallback:
return UnsupportedCloudASRClient(providerId: providerId)
}
}
}
// MARK: - Zhipu (hotwords + prompt)
struct ZhipuCloudASRClient: CloudASRTranscribing {
let apiKey: String
let model: String
let session: URLSession
private static let maxDurationSeconds: TimeInterval = 30
func prepare(dictionary: PersonalDictionary) async throws {}
func transcribe(
samples: [Float],
sampleRate: Int,
locale: Locale,
dictionary: PersonalDictionary
) async throws -> String {
guard !apiKey.isEmpty else { throw CloudASRError.noAPIKey }
let duration = Double(samples.count) / Double(sampleRate)
guard duration <= Self.maxDurationSeconds else { throw CloudASRError.audioTooLong }
let wav = PCMSampleWavEncoder.encode(samples: samples, sampleRate: sampleRate)
let urlString = "https://open.bigmodel.cn/api/paas/v4\(CloudASRModelCatalog.zhipuTranscriptionPath)"
guard let url = URL(string: urlString) else { throw CloudASRError.invalidURL }
let boundary = "Boundary-\(UUID().uuidString)"
var body = Data()
func appendField(_ name: String, _ value: String) {
body.append("--\(boundary)\r\n".data(using: .utf8)!)
body.append("Content-Disposition: form-data; name=\"\(name)\"\r\n\r\n".data(using: .utf8)!)
body.append("\(value)\r\n".data(using: .utf8)!)
}
appendField("model", model)
appendField("stream", "false")
let hotwords = dictionary.asrHotwords()
if !hotwords.isEmpty,
let hotwordsJSON = try? JSONSerialization.data(withJSONObject: hotwords),
let hotwordsString = String(data: hotwordsJSON, encoding: .utf8) {
appendField("hotwords", hotwordsString)
}
let prompt = dictionary.asrPromptBias()
if !prompt.isEmpty {
appendField("prompt", prompt)
}
body.append("--\(boundary)\r\n".data(using: .utf8)!)
body.append("Content-Disposition: form-data; name=\"file\"; filename=\"chunk.wav\"\r\n".data(using: .utf8)!)
body.append("Content-Type: audio/wav\r\n\r\n".data(using: .utf8)!)
body.append(wav)
body.append("\r\n".data(using: .utf8)!)
body.append("--\(boundary)--\r\n".data(using: .utf8)!)
var request = URLRequest(url: url)
request.httpMethod = "POST"
request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
request.setValue("multipart/form-data; boundary=\(boundary)", forHTTPHeaderField: "Content-Type")
request.httpBody = body
request.timeoutInterval = 60
let (data, response) = try await session.data(for: request)
try Self.validateHTTP(response: response, data: data)
guard let text = Self.parseZhipuText(from: data)?.trimmingCharacters(in: .whitespacesAndNewlines),
!text.isEmpty else {
throw CloudASRError.emptyTranscript
}
return text
}
private static func parseZhipuText(from data: Data) -> String? {
guard let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any] else {
return nil
}
return json["text"] as? String
}
fileprivate static func validateHTTP(response: URLResponse, data: Data) throws {
guard let http = response as? HTTPURLResponse else {
throw CloudASRError.transport("non-HTTP response")
}
guard (200..<300).contains(http.statusCode) else {
let message = parseErrorMessage(from: data)
throw CloudASRError.http(status: http.statusCode, message: message)
}
}
fileprivate static func parseErrorMessage(from data: Data) -> String? {
guard let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any] else {
return nil
}
if let message = json["message"] as? String { return message }
if let error = json["error"] as? [String: Any],
let message = error["message"] as? String {
return message
}
return nil
}
}
// MARK: - Alibaba Fun-ASR Flash (HTTP sync, context text bias)
struct AlibabaFunASRClient: CloudASRTranscribing {
let apiKey: String
let model: String
let session: URLSession
func prepare(dictionary: PersonalDictionary) async throws {}
func transcribe(
samples: [Float],
sampleRate: Int,
locale: Locale,
dictionary: PersonalDictionary
) async throws -> String {
guard !apiKey.isEmpty else { throw CloudASRError.noAPIKey }
let dataURI = PCMSampleWavEncoder.dataURI(samples: samples, sampleRate: sampleRate)
let urlString = CloudASRModelCatalog.alibabaAPIBase + CloudASRModelCatalog.alibabaMultimodalPath
guard let url = URL(string: urlString) else { throw CloudASRError.invalidURL }
var messages: [[String: Any]] = []
let context = dictionary.alibabaContextText()
if !context.isEmpty {
messages.append([
"role": "user",
"content": [
["type": "input_text", "text": context]
]
])
}
messages.append([
"role": "user",
"content": [
[
"type": "input_audio",
"input_audio": ["data": dataURI]
]
]
])
let parameters: [String: Any] = [
"format": "wav",
"sample_rate": "\(sampleRate)"
]
let body: [String: Any] = [
"model": model,
"input": ["messages": messages],
"parameters": parameters
]
var request = URLRequest(url: url)
request.httpMethod = "POST"
request.setValue("application/json", forHTTPHeaderField: "Content-Type")
request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
request.setValue("disable", forHTTPHeaderField: "X-DashScope-SSE")
request.httpBody = try JSONSerialization.data(withJSONObject: body)
request.timeoutInterval = 90
let (data, response) = try await session.data(for: request)
try ZhipuCloudASRClient.validateHTTP(response: response, data: data)
if let text = Self.parseText(from: data)?.trimmingCharacters(in: .whitespacesAndNewlines),
!text.isEmpty {
return text
}
return ""
}
private static func parseText(from data: Data) -> String? {
guard let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any],
let output = json["output"] as? [String: Any] else {
return nil
}
if let text = output["text"] as? String { return text }
if let sentence = output["sentence"] as? [String: Any],
let text = sentence["text"] as? String {
return text
}
return nil
}
}
// MARK: - Prompt-biased transcription (OpenAI / MiMo / Groq / custom)
enum PromptCloudASRRequestFormat: Sendable {
case multipart
/// OpenRouter expects JSON `{ model, input_audio: { data, format } }`.
case openRouterJson
}
struct PromptCloudASRClient: CloudASRTranscribing {
let providerId: String
let baseURL: String
let apiKey: String
let model: String
let session: URLSession
var requestFormat: PromptCloudASRRequestFormat = .multipart
/// Groq / OpenRouter batch uploads cap around 30 s per request.
private static let whisperCompatibleMaxDurationSeconds: TimeInterval = 30
func prepare(dictionary: PersonalDictionary) async throws {}
func transcribe(
samples: [Float],
sampleRate: Int,
locale: Locale,
dictionary: PersonalDictionary
) async throws -> String {
guard !apiKey.isEmpty else { throw CloudASRError.noAPIKey }
if providerId == "mimo" {
return try await transcribeMiMo(
samples: samples,
sampleRate: sampleRate,
dictionary: dictionary
)
}
if requestFormat == .openRouterJson {
return try await transcribeOpenRouterJSON(
samples: samples,
sampleRate: sampleRate,
dictionary: dictionary
)
}
return try await transcribeOpenAIStyle(
samples: samples,
sampleRate: sampleRate,
dictionary: dictionary
)
}
private func enforceWhisperDuration(samples: [Float], sampleRate: Int) throws {
let duration = Double(samples.count) / Double(sampleRate)
guard duration <= Self.whisperCompatibleMaxDurationSeconds else {
throw CloudASRError.audioTooLong
}
}
private func transcribeOpenAIStyle(
samples: [Float],
sampleRate: Int,
dictionary: PersonalDictionary
) async throws -> String {
if providerId == "groq" || providerId == "openai" || providerId == "custom" {
try enforceWhisperDuration(samples: samples, sampleRate: sampleRate)
}
let wav = PCMSampleWavEncoder.encode(samples: samples, sampleRate: sampleRate)
let trimmedBase = baseURL.hasSuffix("/") ? String(baseURL.dropLast()) : baseURL
let urlString = "\(trimmedBase)/audio/transcriptions"
guard let url = URL(string: urlString) else { throw CloudASRError.invalidURL }
let boundary = "Boundary-\(UUID().uuidString)"
var body = Data()
func appendField(_ name: String, _ value: String) {
body.append("--\(boundary)\r\n".data(using: .utf8)!)
body.append("Content-Disposition: form-data; name=\"\(name)\"\r\n\r\n".data(using: .utf8)!)
body.append("\(value)\r\n".data(using: .utf8)!)
}
appendField("model", model)
let prompt = dictionary.asrPromptBias(maxCharacters: 600)
if !prompt.isEmpty {
appendField("prompt", prompt)
}
body.append("--\(boundary)\r\n".data(using: .utf8)!)
body.append("Content-Disposition: form-data; name=\"file\"; filename=\"chunk.wav\"\r\n".data(using: .utf8)!)
body.append("Content-Type: audio/wav\r\n\r\n".data(using: .utf8)!)
body.append(wav)
body.append("\r\n".data(using: .utf8)!)
body.append("--\(boundary)--\r\n".data(using: .utf8)!)
var request = URLRequest(url: url)
request.httpMethod = "POST"
request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
request.setValue("multipart/form-data; boundary=\(boundary)", forHTTPHeaderField: "Content-Type")
request.httpBody = body
request.timeoutInterval = 90
let (data, response) = try await session.data(for: request)
try ZhipuCloudASRClient.validateHTTP(response: response, data: data)
guard let text = Self.parseOpenAIText(from: data)?.trimmingCharacters(in: .whitespacesAndNewlines),
!text.isEmpty else {
throw CloudASRError.emptyTranscript
}
return text
}
private func transcribeOpenRouterJSON(
samples: [Float],
sampleRate: Int,
dictionary: PersonalDictionary
) async throws -> String {
try enforceWhisperDuration(samples: samples, sampleRate: sampleRate)
let wav = PCMSampleWavEncoder.encode(samples: samples, sampleRate: sampleRate)
let trimmedBase = baseURL.hasSuffix("/") ? String(baseURL.dropLast()) : baseURL
let urlString = "\(trimmedBase)/audio/transcriptions"
guard let url = URL(string: urlString) else { throw CloudASRError.invalidURL }
var body: [String: Any] = [
"model": model,
"input_audio": [
"data": wav.base64EncodedString(),
"format": "wav"
]
]
let prompt = dictionary.asrPromptBias(maxCharacters: 600)
if !prompt.isEmpty {
body["prompt"] = prompt
}
var request = URLRequest(url: url)
request.httpMethod = "POST"
request.setValue("application/json", forHTTPHeaderField: "Content-Type")
request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
request.httpBody = try JSONSerialization.data(withJSONObject: body)
request.timeoutInterval = 90
let (data, response) = try await session.data(for: request)
try ZhipuCloudASRClient.validateHTTP(response: response, data: data)
guard let text = Self.parseOpenAIText(from: data)?.trimmingCharacters(in: .whitespacesAndNewlines),
!text.isEmpty else {
throw CloudASRError.emptyTranscript
}
return text
}
private func transcribeMiMo(
samples: [Float],
sampleRate: Int,
dictionary: PersonalDictionary
) async throws -> String {
let dataURI = PCMSampleWavEncoder.dataURI(samples: samples, sampleRate: sampleRate)
let trimmedBase = baseURL.hasSuffix("/") ? String(baseURL.dropLast()) : baseURL
let urlString = "\(trimmedBase)/chat/completions"
guard let url = URL(string: urlString) else { throw CloudASRError.invalidURL }
// MiMo-V2.5-ASR is an end-to-end ASR model: the `content` array must
// hold the single `input_audio` part and nothing else. Adding a text
// prompt (personal-dictionary bias) makes the API reject the request
// with HTTP 400 "param incorrect", so we never send prompt bias here.
let userContent: [[String: Any]] = [
[
"type": "input_audio",
"input_audio": ["data": dataURI]
]
]
let body: [String: Any] = [
"model": model,
"messages": [
["role": "user", "content": userContent]
],
"asr_options": ["language": "auto"]
]
var request = URLRequest(url: url)
request.httpMethod = "POST"
request.setValue("application/json", forHTTPHeaderField: "Content-Type")
request.setValue(apiKey, forHTTPHeaderField: "api-key")
request.httpBody = try JSONSerialization.data(withJSONObject: body)
request.timeoutInterval = 90
let (data, response) = try await session.data(for: request)
try ZhipuCloudASRClient.validateHTTP(response: response, data: data)
guard let text = Self.parseChatCompletionText(from: data)?.trimmingCharacters(in: .whitespacesAndNewlines),
!text.isEmpty else {
throw CloudASRError.emptyTranscript
}
return text
}
private static func parseOpenAIText(from data: Data) -> String? {
guard let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any] else {
return nil
}
return json["text"] as? String
}
private static func parseChatCompletionText(from data: Data) -> String? {
guard let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any],
let choices = json["choices"] as? [[String: Any]],
let first = choices.first,
let message = first["message"] as? [String: Any] else {
return nil
}
return message["content"] as? String
}
}
// MARK: - Unsupported hosted ASR (Moonshot)
struct UnsupportedCloudASRClient: CloudASRTranscribing {
let providerId: String
func prepare(dictionary: PersonalDictionary) async throws {}
func transcribe(
samples: [Float],
sampleRate: Int,
locale: Locale,
dictionary: PersonalDictionary
) async throws -> String {
throw CloudASRError.providerUnsupported
}
}