deepvoice_client/local_plugins/azure_speech/ios/Classes/AzureAsrHelper.swift

import Foundation
import AVFoundation
import MicrosoftCognitiveServicesSpeech
// 自定义语音处理组件，提供音频流处理等功能
import speech
import os.log


/**
 * Azure ASR Helper
 * 
 * 基于微软Azure语音服务的ASR实现
 * 参考文档: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-recognize-speech
 */
public class AzureAsrHelper: NSObject {
    private let tag = "AzureAsrHelper"
    // 日志对象
    private let log = OSLog(subsystem: "com.azure.speech", category: "AzureAsrHelper")
    
    // 核心组件
    private var speechConfig: SPXSpeechConfiguration?
    private var recognizer: SPXSpeechRecognizer?
    private var audioConfig: SPXAudioConfiguration?
    
    // 状态管理
    private var _isContinuousRecognitionActive = false
    
    // 配置参数
    private var currentLanguage = "zh-CN"
    private var supportedLanguages = ["zh-CN"]
    private var isAutoDetectLanguage = false
    private var subscriptionKey = ""
    private var region = ""
    
    // 音频源配置
    public enum AudioSourceType {
        /** 使用设备麦克风 */
        case microphone
        
        /** 使用外部提供的音频数据 */
        case external
    }
    
    private var audioSourceType = AudioSourceType.microphone
    
    // 音频处理
    private var externalAudioStream: ExternalAudioPullStream?
    
    /**
     * 初始化Azure语音服务
     * 
     * @param subscriptionKey Azure 订阅密钥
     * @param region Azure 区域
     * @param supportedLanguages 支持的语言数组
     * @param audioSourceType 音频源类型
     * @return 初始化是否成功
     */
    public func initialize(
        subscriptionKey: String,
        region: String,
        supportedLanguages: [String] = ["zh-CN"],
        audioSourceType: AudioSourceType = .microphone
    ) -> Bool {
        do {
            // 检查配置是否为空
            if subscriptionKey.isEmpty || region.isEmpty {
                os_log("Azure 配置信息不完整", log: log, type: .error)
                return false
            }
            
            // 释放之前的资源
            dispose()
            
            // 保存配置
            self.subscriptionKey = subscriptionKey
            self.region = region
            self.audioSourceType = audioSourceType
            
            // 设置语言
            if !supportedLanguages.isEmpty {
                self.supportedLanguages = supportedLanguages
            }
            
            // 根据支持的语言数量决定是否启用自动语言检测
            self.isAutoDetectLanguage = supportedLanguages.count >= 2
            
            // 如果只有一种语言，设置为当前语言
            if !isAutoDetectLanguage && !supportedLanguages.isEmpty {
                self.currentLanguage = supportedLanguages[0]
            }
            
            // 创建语音配置
            speechConfig = try SPXSpeechConfiguration(subscription: subscriptionKey, region: region)
            
            if isAutoDetectLanguage {
                // 直接启用语言检测模式
                speechConfig?.setPropertyTo("Continuous", byName: "SpeechServiceConnection_LanguageIdMode")
                os_log("启用语言检测模式", log: log, type: .info)
            } else {
                // 设置指定的识别语言
                speechConfig?.speechRecognitionLanguage = currentLanguage
            }
            
            // 创建识别器
            return setupRecognizer()
        } catch {
            os_log("初始化失败: %{public}@", log: log, type: .error, error.localizedDescription)
            return false
        }
    }
    
    /**
     * 向音频流写入音频数据
     * 仅当音频源设置为external时有效
     * 
     * @param data 音频数据字节数组
     */
    public func pushAudioData(data: Data) {
        if audioSourceType != .external {
            return
        }
        
        // 使用拉流模式，将数据推入队列
        externalAudioStream?.pushAudio(data)
    }
    
    /**
     * 开始连续语音识别
     * 
     * @param callback 连续识别结果回调
     * @param audioSourceType 音频源类型
     * @return 是否成功开始识别
     */
    public func startContinuousRecognition(
        callback: ContinuousRecognizeCallback,
        audioSourceType: AudioSourceType = .microphone
    ) -> Bool {
        guard speechConfig != nil else {
            callback.onError("语音服务未初始化")
            return false
        }
        
        if _isContinuousRecognitionActive {
            return true
        }
        
        self.audioSourceType = audioSourceType
        
        // 重置识别器
        if !setupRecognizer() {
            callback.onError("重置识别器失败")
            return false
        }
        
        do {
            // 设置各种事件监听
            setupEventListeners(callback: callback)
            
            // 启动音频处理
            startAudioProcessing()
            
            // 开始连续识别
            try recognizer?.startContinuousRecognition()
            _isContinuousRecognitionActive = true
            
            os_log("连续识别已启动，音频源: %{public}@", log: log, type: .info, audioSourceType == .microphone ? "麦克风" : "外部")
            
            return true
        } catch {
            stopAudioProcessing()
            _isContinuousRecognitionActive = false
            callback.onError("启动连续识别失败: \(error.localizedDescription)")
            return false
        }
    }
    
    /**
     * 停止连续语音识别
     *
     * @return 是否成功停止
     */
    public func stopContinuousRecognition() -> Bool {
        guard speechConfig != nil else {
            os_log("语音服务未初始化", log: log, type: .error)
            return false
        }
        
        if !_isContinuousRecognitionActive {
            return true
        }
        
        do {
            guard let recognizer = recognizer else {
                os_log("识别器为空，重置状态", log: log, type: .info)
                _isContinuousRecognitionActive = false
                return true
            }
            
            if audioSourceType == .external {
                pushAudioData(data: Data())
            }
            
            // 停止连续识别
            try recognizer.stopContinuousRecognition()
            
            // 停止音频处理
            stopAudioProcessing()
            
            // 会话结束事件会设置_isContinuousRecognitionActive = false
            return true
        } catch {
            // 强制重置状态
            _isContinuousRecognitionActive = false
            os_log("停止连续识别失败: %{public}@", log: log, type: .error, error.localizedDescription)
            
            // 停止音频处理
            stopAudioProcessing()
            
            // 尝试强制关闭识别器
            recognizer = nil
            
            return false
        }
    }
    
    /**
     * 检查连续识别是否活跃
     */
    public func isContinuousRecognitionActive() -> Bool {
        return self._isContinuousRecognitionActive
    }
    
    /**
     * 释放所有资源
     */
    public func dispose() {
        do {
            // 如果正在进行连续识别，先停止
            if _isContinuousRecognitionActive {
                // 直接停止，不等待结果
                try? recognizer?.stopContinuousRecognition()
                _isContinuousRecognitionActive = false
            }
            
            // 停止音频处理
            stopAudioProcessing()
            
            // 释放资源
            recognizer = nil
            speechConfig = nil
            audioConfig = nil
            
            // 确保状态被重置
            _isContinuousRecognitionActive = false
            externalAudioStream = nil
        } catch {
            // 确保状态被重置
            _isContinuousRecognitionActive = false
            externalAudioStream = nil
            audioConfig = nil
            recognizer = nil
            speechConfig = nil
        }
    }
    
    // MARK: - 私有方法
    
    /**
     * 设置识别器
     */
    private func setupRecognizer() -> Bool {
        do {
            // 清理旧的识别器
            recognizer = nil
            
            // 设置音频配置
            switch audioSourceType {
            case .microphone:
                // 使用默认麦克风输入配置
                audioConfig = try SPXAudioConfiguration()
            case .external:
                // 使用拉流方式处理外部音频
                setupExternalAudioStream()
            }
            
            // 创建识别器
            if isAutoDetectLanguage {
                let autoDetectConfig = try SPXAutoDetectSourceLanguageConfiguration(supportedLanguages)
                recognizer = try SPXSpeechRecognizer(
                    speechConfiguration: speechConfig!,
                    autoDetectSourceLanguageConfiguration: autoDetectConfig,
                    audioConfiguration: audioConfig!
                )
            } else {
                recognizer = try SPXSpeechRecognizer(
                    speechConfiguration: speechConfig!,
                    audioConfiguration: audioConfig!
                )
            }
            
            return true
        } catch {
            os_log("创建识别器失败: %{public}@", log: log, type: .error, error.localizedDescription)
            stopAudioProcessing()
            return false
        }
    }
    
    /**
     * 设置外部音频流 - 使用拉流方式
     */
    private func setupExternalAudioStream() {
        do {
            // 创建外部音频拉流对象
            externalAudioStream = ExternalAudioPullStream()
            
            // 创建音频配置
            audioConfig = SPXAudioConfiguration(streamInput: externalAudioStream!.pullStream)
        } catch {
            os_log("设置外部音频流失败: %{public}@", log: log, type: .error, error.localizedDescription)
            externalAudioStream = nil
        }
    }
    
    /**
     * 设置事件监听器
     */
    private func setupEventListeners(callback: ContinuousRecognizeCallback) {
        guard let recognizer = recognizer else { return }
        
        // 识别中事件
        recognizer.addRecognizingEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            
            let result = event.result
            let detectedLanguage = self.getDetectedLanguage(from: result)
            // os_log("识别中: %{public}@", log: self.log, type: .info, result.text ?? "")
            // 直接在当前线程调用回调
            callback.onRecognizing(result.text ?? "", detectedLanguage)
        }
        
        // 识别完成事件
        recognizer.addRecognizedEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            
            let result = event.result
            if result.reason == SPXResultReason.recognizedSpeech {
                let detectedLanguage = self.getDetectedLanguage(from: result)
                // 直接在当前线程调用回调
                callback.onResult(result.text ?? "", detectedLanguage)
            }
        }
        
        // 会话开始事件
        recognizer.addSessionStartedEventHandler { (sender, event) in
            // 直接在当前线程调用回调
            callback.onSessionStarted()
        }
        
        // 会话结束事件
        recognizer.addSessionStoppedEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            
            // 直接在当前线程调用回调
            callback.onSessionStopped()
            self._isContinuousRecognitionActive = false
            self.stopAudioProcessing()
        }
        
        // 取消事件
        recognizer.addCanceledEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            
            let errorDetails = event.errorDetails ?? "未知错误"
            let reason = String(describing: event.reason.rawValue)
            
            os_log("识别取消: %{public}@", log: self.log, type: .error, errorDetails)
            
            callback.onCanceled(reason, errorDetails)
        }
    }
    
    /**
     * 获取检测到的语言
     */
    private func getDetectedLanguage(from result: SPXSpeechRecognitionResult) -> String {
        if !isAutoDetectLanguage {
            return ""
        }
        
        // 尝试从属性中获取语言
        if let properties = result.properties,
           let language = properties.getPropertyByName("SpeechServiceResponse_RecognitionLanguage") {
            return language
        }
        
        // 尝试另一种方式获取语言
        do {
            let langResult = try SPXAutoDetectSourceLanguageResult(result)
            return langResult.language ?? ""
        } catch {
            return ""
        }
    }
    
    /**
     * 启动音频处理
     */
    private func startAudioProcessing() {
        switch audioSourceType {
        case .microphone:
            // 拉流模式不需要额外启动，SDK会自动拉取数据
            break
        case .external:
            // 外部音频数据模式下不需要启动处理，等待外部调用pushAudioData
            break
        }
    }
    
    /**
     * 停止音频处理
     */
    private func stopAudioProcessing() {
        if let stream = externalAudioStream {
            stream.close()
            externalAudioStream = nil
            // os_log("外部音频流已关闭", log: log, type: .info)
        }
    }
  
    
    /**
     * 外部音频拉流
     * 实现PullAudioInputStreamCallback，将外部推送的音频数据转换为SDK可拉取的形式
     */
    private class ExternalAudioPullStream: NSObject {
        private(set) var pullStream: SPXPullAudioInputStream!
        private let queue = LinkedBlockingQueue<Data>()
        private var closed = false
        
        override init() {
            super.init()
            
            pullStream = SPXPullAudioInputStream(
                readHandler: { [weak self] (data: NSMutableData, size: UInt) -> Int in
                    guard let self = self else { return 0 }
                    return self.read(buffer: data, size: Int(size))
                },
                closeHandler: { [weak self] in
                    self?.close()
                }
            )
        }
        
        /**
         * 外部调用：推送音频数据到队列
         * @param data 音频数据
         */
        func pushAudio(_ data: Data) {
            if !closed {
                queue.put(data)
            }
        }
        
        /**
         * SDK调用：从队列中拉取数据
         * @param buffer SDK提供的缓冲区
         * @param size 缓冲区大小
         * @return 读取的字节数，0表示流结束
         */
        private func read(buffer: NSMutableData, size: Int) -> Int {
            // 阻塞等待下一块数据
            guard let chunk = queue.take() else {
                return 0 // 队列已关闭
            }
            
            // 如果是空数据，表示流结束
            if chunk.isEmpty {
                return 0
            }
            
            let toCopy = min(chunk.count, size)
            buffer.append(chunk.prefix(toCopy))
            return toCopy
        }
        
        /**
         * SDK调用：关闭流
         */
        func close() {
            closed = true
            queue.close()
        }
    }
    
    /**
     * iOS版LinkedBlockingQueue实现
     * 与Android LinkedBlockingQueue保持一致的API和行为
     */
    private class LinkedBlockingQueue<T> {
        private var queue: [T] = []
        private var isClosed = false
        
        // 使用DispatchSemaphore实现阻塞行为
        private let availableItems: DispatchSemaphore
        private let queueLock = NSLock()
        
        init() {
            self.availableItems = DispatchSemaphore(value: 0)
        }
        
        /**
         * 阻塞式获取元素（等价于Android的take()）
         * @return 队列中的元素，如果队列已关闭则返回nil
         */
        func take() -> T? {
            // 等待可用元素（阻塞直到有元素或队列关闭）
            availableItems.wait()
            
            queueLock.lock()
            defer { queueLock.unlock() }
            
            // 检查队列是否已关闭
            if isClosed && queue.isEmpty {
                return nil
            }
            
            // 获取第一个元素
            guard !queue.isEmpty else {
                return nil
            }
            
            return queue.removeFirst()
        }
        
        /**
         * 添加元素（等价于Android的put()）
         * @param item 要添加的元素
         */
        func put(_ item: T) {
            queueLock.lock()
            defer { queueLock.unlock() }
            
            // 检查队列是否已关闭
            if isClosed {
                return
            }
            
            // 添加元素（无容量限制，与Android一致）
            queue.append(item)
            
            // 通知有新元素可用
            availableItems.signal()
        }
        
        /**
         * 关闭队列
         * 关闭后不能再添加新元素，但可以继续取出已有元素
         */
        func close() {
            queueLock.lock()
            defer { queueLock.unlock() }
            
            isClosed = true
            
            // 唤醒所有等待的take()操作
            for _ in 0..<100 { // 假设最多100个等待者
                availableItems.signal()
            }
        }
    }
    
    /**
     * 连续识别回调接口
     */
    public protocol ContinuousRecognizeCallback {
        /**
         * 返回识别结果
         * 
         * @param text 识别的文本
         * @param detectedLanguage 检测到的语言
         */
        func onResult(_ text: String, _ detectedLanguage: String)
        
        /**
         * 识别进行中调用
         * 
         * @param recognizing 正在识别的文本
         * @param detectedLanguage 检测到的语言
         */
        func onRecognizing(_ recognizing: String, _ detectedLanguage: String)
        
        /**
         * 会话开始时调用
         */
        func onSessionStarted()
        
        /**
         * 会话结束时调用
         */
        func onSessionStopped()
        
        /**
         * 识别取消时调用
         * 
         * @param reason 取消原因
         * @param errorDetails 错误详情
         */
        func onCanceled(_ reason: String, _ errorDetails: String)
        
        /**
         * 识别出错时调用
         * 
         * @param error 错误信息
         */
        func onError(_ error: String)
    }
}