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

/**
 * 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
 */
class AzureAsrHelper: NSObject, IAsrService {
    private let tag = "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
    
    // 音频源配置
    private 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: ExternalAudioStream?
    
    // 事件监听处理
    private var eventListeners = NSHashTable<AnyObject>.weakObjects()
    
    // 音频会话
    private let audioSession = AVAudioSession.sharedInstance()
    
    /**
     * 初始化语音识别服务
     * 
     * @param subscriptionKey 服务订阅密钥/访问令牌
     * @param region 服务区域/应用ID
     * @param supportedLanguages 支持的语言列表，如["zh-CN", "en-US"]
     * @return 是否初始化成功
     */
    func initialize(subscriptionKey: String, region: String, supportedLanguages: [String]?) -> Bool {
        do {
            // 检查配置是否为空
            if subscriptionKey.isEmpty || region.isEmpty {
                print("\(tag): Azure 配置信息不完整")
                return false
            }
            
            // 释放之前的资源
            dispose()
            
            // 设置语言
            if let supportedLanguages = supportedLanguages, !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 = self.supportedLanguages.count >= 2
            
            // 如果只有一种语言，设置为当前语言
            if !isAutoDetectLanguage && !self.supportedLanguages.isEmpty {
                self.currentLanguage = self.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")
                print("AzureAsrHelper: 启用语言检测模式")
                
            } else {
                // 设置指定的识别语言
                speechConfig?.speechRecognitionLanguage = currentLanguage
            }
            
            // 创建识别器
            return setupRecognizer()
        } catch {
            print("\(tag): 初始化失败: \(error.localizedDescription)")
            return false
        }
    }
    
    /**
     * 单次识别（非流式）
     * 
     * @param audioData 要识别的音频数据字节数组
     * @param callback 识别结果回调
     * @return 是否成功开始识别
     */
    func recognizeOnce(audioData: Data, callback: @escaping (String) -> Void) -> Bool {
        return false
    }
    
    /**
     * 启动连续识别
     * 
     * @return 是否成功启动
     */
    func startContinuousRecognition() -> Bool {
        if isContinuousRecognitionActive {
            return true
        }
        
        guard let recognizer = recognizer else {
            print("\(tag): 启动连续识别失败: 识别器未初始化")
            return false
        }
        
        do {
            // 启动连续识别
            try recognizer.startContinuousRecognition()
            isContinuousRecognitionActive = true
            return true
        } catch {
            print("\(tag): 启动连续识别失败: \(error.localizedDescription)")
            return false
        }
    }
    
    /**
     * 停止连续识别
     * 
     * @return 是否成功停止
     */
    func stopContinuousRecognition() -> Bool {
        guard let recognizer = recognizer, isContinuousRecognitionActive else {
            return true
        }
        
        do {
            // 停止连续识别
            try recognizer.stopContinuousRecognition()
            isContinuousRecognitionActive = false
            return true
        } catch {
            print("\(tag): 停止连续识别失败: \(error.localizedDescription)")
            return false
        }
    }
    
    /**
     * 推送音频数据
     * 在连续识别模式下推送新的音频数据
     * 
     * @param audioData 要识别的音频数据字节数组
     * @return 是否成功处理
     */
    func pushAudioData(audioData: Data) -> Bool {
        if !isContinuousRecognitionActive {
            return false
        }
        
        if audioSourceType != .external {
            notifyEvent(eventType: .error, params: ["error": "当前不是外部音频流模式"])
            return false
        }
        
        // 将音频数据推送到外部音频流
        externalAudioStream?.pushAudio(audioData)
        return true
    }
    
    /**
     * 添加ASR事件监听器
     * 
     * @param listener 事件监听器
     */
    func addListener(_ listener: AsrEventListener) {
        eventListeners.add(listener as AnyObject)
    }
    
    /**
     * 移除ASR事件监听器
     * 
     * @param listener 要移除的事件监听器
     */
    func removeListener(_ listener: AsrEventListener) {
        eventListeners.remove(listener as AnyObject)
    }
    
    /**
     * 释放资源
     * 在不再需要服务时调用，释放底层资源
     */
    func dispose() {
        // 停止识别
        if isContinuousRecognitionActive {
            _ = stopContinuousRecognition()
        }
        
        // 停止音频处理
        stopAudioProcessing()
        
        // 释放资源
        recognizer = nil
        speechConfig = nil
        audioConfig = nil
        externalAudioStream = nil
    }
    
    /**
     * 设置识别参数
     * 
     * @param params 参数键值对
     * @return 是否设置成功
     */
    func setParameters(_ params: [String: Any]) -> Bool {
        // 未实现自定义参数，后续可以根据需要添加
        return true
    }
    
    /**
     * 获取连续识别是否活跃
     */
    func isRecognitionActive() -> Bool {
        return isContinuousRecognitionActive
    }
    
    // MARK: - 辅助方法
    
    /**
     * 创建识别器
     * 
     * @return 是否创建成功
     */
    private func setupRecognizer() -> Bool {
        guard let speechConfig = speechConfig else {
            print("\(tag): 无法创建识别器：语音配置不存在")
            return false
        }
        
        do {
            if audioSourceType == .external {
                // 设置外部音频流
                setupExternalAudioStream()
            }
            
            // 根据是否启用自动识别语言创建不同的识别器
            if isAutoDetectLanguage && supportedLanguages.count > 0 {
                // 创建语言自动检测配置
                let autoDetectConfig = try SPXAutoDetectSourceLanguageConfiguration(supportedLanguages)
                
                if let audioConfig = self.audioConfig {
                    // 使用指定的音频源创建识别器
                    recognizer = try SPXSpeechRecognizer(speechConfiguration: speechConfig, 
                                                        autoDetectSourceLanguageConfiguration: autoDetectConfig,
                                                        audioConfiguration: audioConfig)
                } else {
                    // 使用默认音频源创建识别器
                    // 创建一个默认的麦克风配置而不是传nil
                    let defaultAudioConfig = SPXAudioConfiguration()
                    recognizer = try SPXSpeechRecognizer(speechConfiguration: speechConfig, 
                                                        autoDetectSourceLanguageConfiguration: autoDetectConfig,
                                                        audioConfiguration: defaultAudioConfig)
                }
            } else {
                // 使用单一语言配置创建识别器
                if let audioConfig = self.audioConfig {
                    // 使用指定的音频源创建识别器
                    recognizer = try SPXSpeechRecognizer(speechConfiguration: speechConfig,
                                                        audioConfiguration: audioConfig)
                } else {
                    // 使用默认音频源创建识别器
                    // 创建一个默认的麦克风配置
                    let defaultAudioConfig = SPXAudioConfiguration()
                    recognizer = try SPXSpeechRecognizer(speechConfiguration: speechConfig,
                                                        audioConfiguration: defaultAudioConfig)
                }
            }
            
            // 设置事件监听器
            setupEventListeners()
            
            return true
        } catch {
            print("\(tag): 创建识别器失败: \(error.localizedDescription)")
            stopAudioProcessing()
            return false
        }
    }
    
    /**
     * 停止音频处理
     */
    private func stopAudioProcessing() {
        // 清理外部音频流
        externalAudioStream = nil
    }
    
    /**
     * 设置外部音频流
     */
    private func setupExternalAudioStream() {
        externalAudioStream = ExternalAudioStream()
        // 使用ExternalAudioStream实例中创建的pullStream
        audioConfig = SPXAudioConfiguration(streamInput: externalAudioStream!.pullStream)
    }
    
    /**
     * 设置事件监听器
     */
    private func setupEventListeners() {
        guard let recognizer = recognizer else { return }
        
        // 修改为直接设置事件回调
        recognizer.addRecognizingEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            
            var detectedLanguage = ""
            // 修改event.result的处理，不使用可选链
            let result = event.result
            if self.isAutoDetectLanguage {
                // 处理自动语言检测结果
                if let properties = result.properties, 
                   let language = properties.getPropertyByName("SpeechServiceResponse_RecognitionLanguage") {
                    detectedLanguage = language
                }
            }
            
            self.notifyEvent(eventType: .recognizing, params: [
                "text": result.text,
                "language": detectedLanguage
            ])
        }
        
        // 添加最终识别结果处理器
        recognizer.addRecognizedEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            
            // 修改event.result的处理，不使用可选绑定
            let result = event.result
            var detectedLanguage = ""
            
            if self.isAutoDetectLanguage {
                // 处理自动语言检测结果
                if let properties = result.properties,
                   let language = properties.getPropertyByName("SpeechServiceResponse_RecognitionLanguage") {
                    detectedLanguage = language
                }
                
                // 尝试另一种方式获取语言
                do {
                    let langResult = try SPXAutoDetectSourceLanguageResult(result)
                    if let lang = langResult.language, !lang.isEmpty {
                        detectedLanguage = lang
                    }
                } catch {
                    // 忽略错误，继续使用之前获取的语言
                }
            }
            
            self.notifyEvent(eventType: .recognized, params: [
                "text": result.text,
                "language": detectedLanguage
            ])
        }
        
        // 添加会话开始事件处理器
        recognizer.addSessionStartedEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            self.notifyEvent(eventType: .sessionStarted)
        }
        
        // 添加会话结束事件处理器
        recognizer.addSessionStoppedEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            self.isContinuousRecognitionActive = false
            self.notifyEvent(eventType: .sessionStopped)
        }
        
        // 添加取消事件处理器
        recognizer.addCanceledEventHandler { [weak self] (sender, event) in
            guard let self = self else { return }
            
            print("\(self.tag): 识别取消: \(event.errorDetails ?? "未知错误")")
            
            // 标记连续识别已停止
            self.isContinuousRecognitionActive = false
            
            // 通知错误
            self.notifyEvent(eventType: .canceled, params: [
                "error": "识别取消",
                "errorDetails": event.errorDetails ?? "未知错误",
                "reason": String(event.reason.rawValue)
            ])
        }
    }
    
    /**
     * 触发事件通知
     */
    private func notifyEvent(eventType: AsrEventType, params: [String: Any] = [:]) {
        let event = AsrEvent(type: eventType, params: params)
        
        // 在主线程通知事件
        DispatchQueue.main.async {
            for case let listener as AsrEventListener in self.eventListeners.allObjects {
                listener.onEvent(event)
            }
        }
    }
    
    /**
     * 外部音频流 - 使用拉流模式
     * 基于NSObject实现，将外部推送的音频数据转换为SDK可拉取的形式
     */
    class ExternalAudioStream: NSObject {
        // 内部持有一个SPXPullAudioInputStream实例
        private(set) var pullStream: SPXPullAudioInputStream!
        
        // 音频数据缓冲区
        private var buffer = Data()
        private let lock = NSLock()
        
        // 记录是否已关闭
        private var isClosed = false
        
        /**
         * 初始化方法
         */
        override init() {
            super.init()
            
            // 创建拉流，使用官方文档推荐的参数类型
            pullStream = SPXPullAudioInputStream(
                readHandler: { [weak self] (data: NSMutableData, size: UInt) -> Int in
                    guard let self = self else { return 0 }
                    self.lock.lock()
                    defer { self.lock.unlock() }
                    if self.buffer.isEmpty { return 0 }
                    let bytesToRead = min(Int(size), self.buffer.count)
                    data.append(self.buffer.prefix(bytesToRead))
                    self.buffer.removeFirst(bytesToRead)
                    return bytesToRead
                },
                closeHandler: { [weak self] in
                    self?.close()
                }
            )
        }
        
        /**
         * 推送音频数据到缓冲区
         * 
         * @param data 要推送的音频数据
         */
        func pushAudio(_ data: Data) {
            if isClosed { return }
            
            lock.lock()
            defer { lock.unlock() }
            
            // 添加到缓冲区
            buffer.append(data)
        }
        
        /**
         * 关闭流
         */
        func close() {
            lock.lock()
            defer { lock.unlock() }
            
            isClosed = true
            buffer = Data()
        }
    }
}