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

import Foundation
import AVFoundation
import MicrosoftCognitiveServicesSpeech

/// Azure语音识别辅助类，支持麦克风和外部音频源的单次和连续语音识别
class AzureAsrHelper {
    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
    private var subscriptionKey = ""
    private var region = ""
    
    // 音频源配置
    private var audioSourceType = AudioSourceType.microphone
    
    // 音频处理
    private var microphoneStream: MicrophoneStream?
    private var externalAudioStream: ExternalAudioPullStream?
    
    // 连续识别回调
    private var continuousRecognizeCallback: ContinuousRecognizeCallback?
    
    /// 应用上下文
    private let context: Any
    
    /// 音频来源类型
    enum AudioSourceType {
        /// 使用设备麦克风
        case microphone
        
        /// 使用外部提供的音频数据
        case external
    }
    
    /// 一次性识别回调
    protocol RecognizeCallback {
        /// 识别结果回调
        /// - Parameters:
        ///   - text: 识别文本
        ///   - detectedLanguage: 检测到的语言
        func onResult(text: String, detectedLanguage: String)
        
        /// 识别错误回调
        /// - Parameter error: 错误信息
        func onError(error: String)
    }
    
    /// 连续识别回调
    protocol ContinuousRecognizeCallback {
        /// 识别中回调
        /// - Parameters:
        ///   - recognizing: 识别中的文本
        ///   - detectedLanguage: 检测到的语言
        func onRecognizing(recognizing: String, detectedLanguage: String)
        
        /// 识别结果回调
        /// - Parameters:
        ///   - text: 识别文本
        ///   - detectedLanguage: 检测到的语言
        func onResult(text: String, detectedLanguage: String)
        
        /// 会话开始回调
        func onSessionStarted()
        
        /// 会话结束回调
        func onSessionStopped()
        
        /// 识别取消回调
        /// - Parameters:
        ///   - reason: 取消原因
        ///   - errorDetails: 错误详情
        func onCanceled(reason: String, errorDetails: String)
        
        /// 识别错误回调
        /// - Parameter error: 错误信息
        func onError(error: String)
    }
    
    /// 初始化Azure语音识别辅助类
    /// - Parameter context: 应用上下文
    init(_ context: Any) {
        self.context = context
    }
    
    /// 初始化Azure语音服务
    /// - Parameters:
    ///   - subscriptionKey: Azure 订阅密钥
    ///   - region: Azure 区域
    ///   - supportedLanguages: 支持的语言数组，默认为["zh-CN"]
    ///   - audioSourceType: 音频源类型，默认为麦克风
    /// - Returns: 初始化是否成功
    func initialize(
        subscriptionKey: String,
        region: String,
        supportedLanguages: [String] = ["zh-CN"],
        audioSourceType: AudioSourceType = .microphone
    ) -> Bool {
        do {
            print("\(tag): 初始化 Azure 语音服务, 音频源类型: \(audioSourceType)")
            
            // 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()
            
            // 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 {
                // 启用语言检测模式
                try speechConfig?.setPropertyTo("Continuous", byId: SPXPropertyId.speechServiceConnection_languageIdMode)
            } else {
                // 设置指定的识别语言
                speechConfig?.speechRecognitionLanguage = currentLanguage
            }
            
            // 设置音频配置
            return setupRecognizer()
            
        } catch {
            print("\(tag): 初始化失败: \(error.localizedDescription)")
            return false
        }
    }
    
    /// 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:
                // 检查是否连接了耳机
                let isHeadsetConnected = isHeadphonesConnected()
                
                if !isHeadsetConnected {
                    print("\(tag): 未检测到耳机，使用麦克风输入")
                    // 使用拉流模式进行回音消除
                    setupMicrophoneStream()
                } else {
                    print("\(tag): 检测到耳机连接，使用默认麦克风输入")
                    // 使用默认麦克风输入
                    audioConfig = SPXAudioConfiguration()
                }
            
            case .external:
                // 创建拉流
                setupExternalAudioStream()
                print("\(tag): 使用外部音频源(拉流模式)")
            }
            
            // 创建识别器
            if isAutoDetectLanguage {
                let autoDetectConfig = try SPXAutoDetectSourceLanguageConfiguration(sourceLanguages: supportedLanguages)
                recognizer = try SPXSpeechRecognizer(speechConfiguration: speechConfig!, autoDetectSourceLanguageConfiguration: autoDetectConfig, audioConfiguration: audioConfig!)
            } else {
                recognizer = try SPXSpeechRecognizer(speechConfiguration: speechConfig!, audioConfiguration: audioConfig!)
            }
            
            print("\(tag): Azure 语音服务初始化成功")
            return true
        } catch {
            print("\(tag): 创建识别器失败: \(error.localizedDescription)")
            stopAudioProcessing()
            return false
        }
    }
    
    /// 设置麦克风流 - 使用拉流方式
    private func setupMicrophoneStream() {
        do {
            // 创建麦克风流
            microphoneStream = MicrophoneStream()
            
            // 创建音频配置
            audioConfig = try SPXAudioConfiguration(streamInput: microphoneStream!)
            
            print("\(tag): 已设置麦克风流（拉流模式）")
        } catch {
            print("\(tag): 设置麦克风流失败: \(error.localizedDescription)")
        }
    }
    
    /// 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 = try SPXAudioConfiguration(streamInput: externalAudioStream!)
            
            print("\(tag): 已设置外部音频流（拉流模式）")
        } catch {
            print("\(tag): 设置外部音频流失败: \(error.localizedDescription)")
        }
    }
    
    /// 向音频流写入音频数据
    /// 仅当音频源设置为external时有效
    /// - Parameter data: 音频数据字节数组
    func pushAudioData(_ data: Data) {
        if audioSourceType != .external {
            print("\(tag): 当前未使用外部音频源，忽略推送的音频数据")
            return
        }
        
        // 使用拉流模式，将数据推入队列
        externalAudioStream?.pushAudio(data)
    }
    
    /// 执行一次性语音识别
    /// - Parameter callback: 识别结果回调
    func recognizeOnce(callback: RecognizeCallback) {
        if speechConfig == nil {
            callback.onError(error: "语音服务未初始化")
            return
        }
        
        // 确保不在连续识别中
        if isContinuousRecognitionActive {
            _ = stopContinuousRecognition()
        }
        
        // 重置识别器
        if !setupRecognizer() {
            callback.onError(error: "重置识别器失败")
            return
        }
        
        // 启动音频处理
        startAudioProcessing()
        
        // 执行同步识别
        do {
            let result = try recognizer?.recognizeOnce()
            
            // 停止音频处理
            stopAudioProcessing()
            
            if let result = result, result.reason == .recognizedSpeech {
                let resultText = result.text ?? ""
                var detectedLanguage = supportedLanguages[0]
                
                // 尝试获取检测到的语言
                if isAutoDetectLanguage, let autoDetectResult = result.properties?.getPropertyByName("SourceLanguage") {
                    detectedLanguage = autoDetectResult
                }
                
                callback.onResult(text: resultText, detectedLanguage: detectedLanguage)
            } else {
                callback.onError(error: "未能识别语音")
            }
        } catch {
            stopAudioProcessing()
            callback.onError(error: "识别异常: \(error.localizedDescription)")
        }
    }
    
    /// 开始连续语音识别
    /// - Parameter callback: 连续识别结果回调
    /// - Returns: 是否成功开始识别
    func startContinuousRecognition(_ callback: ContinuousRecognizeCallback) -> Bool {
        if speechConfig == nil {
            callback.onError(error: "语音服务未初始化")
            return false
        }
        
        if isContinuousRecognitionActive {
            return true
        }
        
        // 重置识别器
        if !setupRecognizer() {
            callback.onError(error: "重置识别器失败")
            return false
        }
        
        // 保存回调
        continuousRecognizeCallback = callback
        
        do {
            // 设置各种事件监听
            setupEventListeners(callback)
            
            // 启动音频处理
            startAudioProcessing()
            
            // 开始连续识别
            try recognizer?.startContinuousRecognition()
            isContinuousRecognitionActive = true
            
            return true
        } catch {
            isContinuousRecognitionActive = false
            stopAudioProcessing()
            callback.onError(error: "启动连续识别失败: \(error.localizedDescription)")
            return false
        }
    }
    
    /// 设置事件监听器
    private func setupEventListeners(_ callback: ContinuousRecognizeCallback) {
        guard let recognizer = recognizer else { return }
        
        // 识别中事件
        recognizer.addRecognizing { [weak self] _, event in
            guard let self = self else { return }
            
            var detectedLanguage = ""
            if self.isAutoDetectLanguage, let lang = event.result.properties?.getPropertyByName("SourceLanguage") {
                detectedLanguage = lang
            }
            
            print("\(self.tag): 识别中: \(event.result.text ?? ""), 语言: \(detectedLanguage)")
            callback.onRecognizing(recognizing: event.result.text ?? "", detectedLanguage: detectedLanguage)
        }
        
        // 识别完成事件
        recognizer.addRecognized { [weak self] _, event in
            guard let self = self else { return }
            
            if event.result.reason == .recognizedSpeech {
                var detectedLanguage = ""
                if self.isAutoDetectLanguage, let lang = event.result.properties?.getPropertyByName("SourceLanguage") {
                    detectedLanguage = lang
                }
                
                print("\(self.tag): 识别完成: \(event.result.text ?? ""), 语言: \(detectedLanguage)")
                callback.onResult(text: event.result.text ?? "", detectedLanguage: detectedLanguage)
            }
        }
        
        // 会话开始事件
        recognizer.addSessionStarted { [weak self] _, _ in
            guard let self = self else { return }
            
            print("\(self.tag): 识别会话已开始")
            callback.onSessionStarted()
        }
        
        // 会话结束事件
        recognizer.addSessionStopped { [weak self] _, _ in
            guard let self = self else { return }
            
            print("\(self.tag): 识别会话已结束")
            self.isContinuousRecognitionActive = false
            self.stopAudioProcessing()
            callback.onSessionStopped()
        }
        
        // 取消事件
        recognizer.addCanceled { [weak self] _, event in
            guard let self = self else { return }
            
            let errorDetails = event.errorDetails ?? "未知错误"
            let reason = String(describing: event.reason)
            
            print("\(self.tag): 识别取消: \(errorDetails)")
            self.isContinuousRecognitionActive = false
            self.stopAudioProcessing()
            
            callback.onCanceled(reason: reason, errorDetails: errorDetails)
            callback.onError(error: "识别取消: \(errorDetails)") // 兼容旧接口
        }
    }
    
    /// 停止连续语音识别
    /// - Returns: 是否成功停止
    func stopContinuousRecognition() -> Bool {
        if speechConfig == nil {
            print("\(tag): 语音服务未初始化")
            return false
        }
        
        if !isContinuousRecognitionActive {
            print("\(tag): 未进行连续识别，忽略停止请求")
            return true
        }
        
        do {
            print("\(tag): 停止连续语音识别")
            
            if recognizer == nil {
                print("\(tag): 识别器为空，重置状态")
                isContinuousRecognitionActive = false
                return true
            }
            
            // 停止连续识别
            try recognizer?.stopContinuousRecognition()
            
            // 停止音频处理
            stopAudioProcessing()
            
            // 等待一段时间以确保资源正确释放
            DispatchQueue.main.asyncAfter(deadline: .now() + 0.5) { [weak self] in
                self?.isContinuousRecognitionActive = false
            }
            
            print("\(tag): 连续识别停止指令已发送")
            return true
        } catch {
            // 强制重置状态
            isContinuousRecognitionActive = false
            print("\(tag): 停止连续识别失败: \(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
    /// - Returns: 是否活跃
    func isContinuousRecognitionActive() -> Bool {
        return isContinuousRecognitionActive
    }
    
    /// 释放所有资源
    func dispose() {
        // 如果正在进行连续识别，先停止
        if isContinuousRecognitionActive {
            _ = stopContinuousRecognition()
        }
        
        // 停止音频处理
        stopAudioProcessing()
        
        // 清理所有资源
        recognizer = nil
        speechConfig = nil
        audioConfig = nil
        microphoneStream = nil
        externalAudioStream = nil
        
        // 重置状态
        isContinuousRecognitionActive = false
        continuousRecognizeCallback = nil
    }
    
    /// 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() {
        microphoneStream?.close()
        microphoneStream = nil
        
        externalAudioStream?.close()
        externalAudioStream = nil
    }
    
    /// 检查是否连接了耳机
    private func isHeadphonesConnected() -> Bool {
        let audioSession = AVAudioSession.sharedInstance()
        let outputs = audioSession.currentRoute.outputs
        
        // 检查是否存在耳机类型的输出
        let headphoneTypes: [AVAudioSession.Port] = [
            .headphones, .bluetoothA2DP, .bluetoothHFP, .bluetoothLE
        ]
        
        for output in outputs {
            if headphoneTypes.contains(output.portType) {
                return true
            }
        }
        
        return false
    }
    
    /// 麦克风流 - 拉流模式
    /// 实现SPXPullAudioInputStreamDelegate，为Azure SDK提供音频数据
    class MicrophoneStream: NSObject, SPXPullAudioInputStreamDelegate {
        private let tag = "MicrophoneStream"
        private var audioEngine: AVAudioEngine?
        private var audioBuffer = Data()
        private let bufferLock = NSLock()
        
        override init() {
            super.init()
            setupAudioEngine()
        }
        
        private func setupAudioEngine() {
            do {
                print("\(tag): 初始化音频引擎")
                
                // 创建音频引擎
                audioEngine = AVAudioEngine()
                
                // 设置音频会话
                let audioSession = AVAudioSession.sharedInstance()
                try audioSession.setCategory(.record, mode: .default)
                try audioSession.setActive(true, options: .notifyOthersOnDeactivation)
                
                // 获取输入节点
                guard let inputNode = audioEngine?.inputNode else {
                    print("\(tag): 无法获取输入节点")
                    return
                }
                
                // 创建16kHz, 16bit, 单声道格式
                let format = AVAudioFormat(commonFormat: .pcmFormatInt16, sampleRate: 16000, channels: 1, interleaved: true)
                
                // 设置音频处理回调
                let bufferSize = 4096
                inputNode.installTap(onBus: 0, bufferSize: UInt32(bufferSize), format: format) { [weak self] (buffer, time) in
                    guard let self = self else { return }
                    
                    // 将音频数据转换为Data
                    let audioBuffer = buffer.int16ChannelData?[0]
                    let audioBufferSize = Int(buffer.frameLength * buffer.format.streamDescription.pointee.mBytesPerFrame)
                    
                    if let audioBuffer = audioBuffer {
                        let data = Data(bytes: audioBuffer, count: audioBufferSize)
                        self.bufferLock.lock()
                        self.audioBuffer.append(data)
                        self.bufferLock.unlock()
                    }
                }
                
                // 启动音频引擎
                audioEngine?.prepare()
                try audioEngine?.start()
                
                print("\(tag): 音频引擎启动成功")
            } catch {
                print("\(tag): 设置音频引擎失败: \(error.localizedDescription)")
            }
        }
        
        // SPXPullAudioInputStreamDelegate协议方法
        func read(_ buffer: UnsafeMutablePointer<UInt8>, size: UInt32) -> Int32 {
            bufferLock.lock()
            defer { bufferLock.unlock() }
            
            let maxSize = Int(size)
            
            if audioBuffer.count == 0 {
                // 没有数据可读
                return 0
            }
            
            let readSize = min(audioBuffer.count, maxSize)
            audioBuffer.copyBytes(to: buffer, from: 0..<readSize)
            
            if readSize < audioBuffer.count {
                // 保留未读的数据
                audioBuffer = audioBuffer.subdata(in: readSize..<audioBuffer.count)
            } else {
                // 清空缓冲区
                audioBuffer = Data()
            }
            
            return Int32(readSize)
        }
        
        // 关闭音频资源
        func close() {
            print("\(tag): 关闭音频引擎")
            
            // 停止音频引擎
            if let inputNode = audioEngine?.inputNode {
                inputNode.removeTap(onBus: 0)
            }
            
            audioEngine?.stop()
            audioEngine = nil
            
            // 关闭音频会话
            do {
                try AVAudioSession.sharedInstance().setActive(false)
            } catch {
                print("\(tag): 关闭音频会话失败: \(error.localizedDescription)")
            }
            
            // 清空缓冲区
            bufferLock.lock()
            audioBuffer = Data()
            bufferLock.unlock()
            
            print("\(tag): 音频资源已释放")
        }
    }
    
    /// 外部音频拉流
    /// 实现SPXPullAudioInputStreamDelegate，将外部推送的音频数据转换为SDK可拉取的形式
    class ExternalAudioPullStream: NSObject, SPXPullAudioInputStreamDelegate {
        private let tag = "ExternalAudioPullStream"
        private let audioQueue = DispatchQueue(label: "com.yunqiinnovation.azure_speech.externalAudio")
        private var audioBuffer = Data()
        private let bufferLock = NSLock()
        private var isClosed = false
        
        // 外部调用：推送音频数据
        func pushAudio(_ data: Data) {
            if isClosed { return }
            
            bufferLock.lock()
            audioBuffer.append(data)
            bufferLock.unlock()
        }
        
        // SPXPullAudioInputStreamDelegate协议方法
        func read(_ buffer: UnsafeMutablePointer<UInt8>, size: UInt32) -> Int32 {
            bufferLock.lock()
            defer { bufferLock.unlock() }
            
            if isClosed {
                return 0
            }
            
            let maxSize = Int(size)
            
            if audioBuffer.count == 0 {
                // 没有数据可读，返回0表示需要等待
                return 0
            }
            
            let readSize = min(audioBuffer.count, maxSize)
            audioBuffer.copyBytes(to: buffer, from: 0..<readSize)
            
            if readSize < audioBuffer.count {
                // 保留未读的数据
                audioBuffer = audioBuffer.subdata(in: readSize..<audioBuffer.count)
            } else {
                // 清空缓冲区
                audioBuffer = Data()
            }
            
            return Int32(readSize)
        }
        
        // 关闭音频资源
        func close() {
            print("\(tag): 关闭外部音频流")
            
            bufferLock.lock()
            isClosed = true
            audioBuffer = Data()
            bufferLock.unlock()
            
            print("\(tag): 外部音频流已关闭")
        }
    }
}