deepvoice_client/azure/ios/Classes/AzureAsrHelper.swift

import Foundation
import MicrosoftCognitiveServicesSpeech
import AVFoundation
import AudioToolbox

/// 自定义麦克风流实现，优化ASR语音输入捕获
class AzureMicrophoneStream: NSObject {
    var ioUnit: AudioUnit?
    var audioFormat: AudioStreamBasicDescription
    var audioBufferList: AudioBufferList
    var audioList: [Float] = []
    let audioListQueue = DispatchQueue(label: "azureAudioListQueue")
    private var isActive = false

    override init() {
        // 音频会话配置
        let audioSession = AVAudioSession.sharedInstance()
        do {
            try audioSession.setCategory(.playAndRecord, 
                                        mode: .voiceChat, 
                                        options: [.allowBluetooth, .defaultToSpeaker, .mixWithOthers])
            try audioSession.setActive(true)
            print("[AzureMicrophoneStream] 音频会话配置成功")
        } catch {
            print("[AzureMicrophoneStream] 配置音频会话失败: \(error.localizedDescription)")
        }
        
        // 设置音频格式为16KHz 16位单声道PCM
        audioFormat = AudioStreamBasicDescription(
            mSampleRate: 16000.0,
            mFormatID: kAudioFormatLinearPCM,
            mFormatFlags: kAudioFormatFlagIsSignedInteger | kAudioFormatFlagIsPacked,
            mBytesPerPacket: 2,
            mFramesPerPacket: 1,
            mBytesPerFrame: 2,
            mChannelsPerFrame: 1,
            mBitsPerChannel: 16,
            mReserved: 0
        )

        audioBufferList = AudioBufferList(
            mNumberBuffers: 1,
            mBuffers: AudioBuffer(
                mNumberChannels: audioFormat.mChannelsPerFrame,
                mDataByteSize: 4096,
                mData: malloc(4096)
            )
        )
        
        super.init()
    }
    
    func start() -> Bool {
        if isActive {
            return true // 已经在运行
        }
        
        if setupAudioUnit() {
            isActive = startAudioUnit()
            return isActive
        }
        return false
    }
    
    private func setupAudioUnit() -> Bool {
        print("[AzureMicrophoneStream] 设置音频单元")

        var ioUnitDescription = AudioComponentDescription(
            componentType: kAudioUnitType_Output,
            componentSubType: kAudioUnitSubType_VoiceProcessingIO,
            componentManufacturer: kAudioUnitManufacturer_Apple,
            componentFlags: 0,
            componentFlagsMask: 0
        )

        guard let ioUnitRef = AudioComponentFindNext(nil, &ioUnitDescription) else {
            print("[AzureMicrophoneStream] 无法找到音频组件")
            return false
        }

        if CheckHasError(AudioComponentInstanceNew(ioUnitRef, &ioUnit), "创建IO单元") {
            ioUnit = nil
            return false
        }

        var enableInput: UInt32 = 1
        let kInputBus: AudioUnitElement = 1
        let kOutputBus: AudioUnitElement = 0
        
        if CheckHasError(AudioUnitSetProperty(ioUnit!, kAudioOutputUnitProperty_EnableIO,
                                             kAudioUnitScope_Input, kInputBus, &enableInput,
                                             UInt32(MemoryLayout<UInt32>.size)), "设置输入总线的EnableIO属性") {
            return false
        }

        var enableOutput: UInt32 = 0
        if CheckHasError(AudioUnitSetProperty(ioUnit!, kAudioOutputUnitProperty_EnableIO,
                                             kAudioUnitScope_Output, kOutputBus,
                                             &enableOutput, UInt32(MemoryLayout<UInt32>.size)), "设置输出总线的EnableIO属性") {
            return false
        }

        var flag: UInt32 = 0
        if CheckHasError(AudioUnitSetProperty(ioUnit!, kAudioUnitProperty_ShouldAllocateBuffer,
                                             kAudioUnitScope_Output, kInputBus, &flag, UInt32(MemoryLayout<UInt32>.size)), "设置ShouldAllocateBuffer属性") {
            return false
        }

        let size = UInt32(MemoryLayout<AudioStreamBasicDescription>.size)
        if CheckHasError(AudioUnitSetProperty(ioUnit!, kAudioUnitProperty_StreamFormat,
                                             kAudioUnitScope_Output, kInputBus, &audioFormat, size), "设置输入总线的StreamFormat属性") {
            return false
        }

        if CheckHasError(AudioUnitSetProperty(ioUnit!, kAudioUnitProperty_StreamFormat,
                                             kAudioUnitScope_Input, kOutputBus, &audioFormat, size), "设置输出总线的StreamFormat属性") {
            return false
        }

        var inputCallback = AURenderCallbackStruct(
            inputProc: AzureMicrophoneStream.OnRecordedDataIsAvailable, 
            inputProcRefCon: UnsafeMutableRawPointer(Unmanaged.passUnretained(self).toOpaque())
        )
        
        if CheckHasError(AudioUnitSetProperty(ioUnit!,
                                             kAudioOutputUnitProperty_SetInputCallback,
                                             kAudioUnitScope_Global, kInputBus,
                                             &inputCallback, UInt32(MemoryLayout<AURenderCallbackStruct>.size)), "设置输入回调") {
            return false
        }

        var hasError = CheckHasError(AudioUnitInitialize(ioUnit!), "初始化IO单元")
        if hasError {
            // 如果初始化失败，重试一次
            Thread.sleep(forTimeInterval: 0.5)
            hasError = CheckHasError(AudioUnitInitialize(ioUnit!), "重试初始化IO单元")
        }

        print("[AzureMicrophoneStream] 音频单元设置\(hasError ? "失败" : "成功")")
        return !hasError
    }

    private func startAudioUnit() -> Bool {
        print("[AzureMicrophoneStream] 启动音频单元")
        guard let ioUnit = ioUnit else {
            print("[AzureMicrophoneStream] IO单元未初始化")
            return false
        }
        return !CheckHasError(AudioOutputUnitStart(ioUnit), "启动IO单元")
    }

    func stop() {
        print("[AzureMicrophoneStream] 停止音频单元")
        guard isActive, let ioUnit = ioUnit else { return }
        
        _ = CheckHasError(AudioOutputUnitStop(ioUnit), "停止IO单元")
        isActive = false
    }

    func dispose() {
        stop()
        
        if let ioUnit = ioUnit {
            _ = CheckHasError(AudioUnitUninitialize(ioUnit), "反初始化IO单元")
            _ = CheckHasError(AudioComponentInstanceDispose(ioUnit), "释放IO单元")
        }
        
        // 释放缓冲区
        if let buffer = audioBufferList.mBuffers.mData {
            free(buffer)
        }
        
        self.ioUnit = nil
        
        // 清空音频数据
        audioListQueue.sync {
            audioList.removeAll()
        }
    }

    static let OnRecordedDataIsAvailable: AURenderCallback = { inRefCon, ioActionFlags, inTimeStamp, inBusNumber, inNumberFrames, ioData in
        let wrapper = Unmanaged<AzureMicrophoneStream>.fromOpaque(inRefCon).takeUnretainedValue()
        let expectedDataByteSize = inNumberFrames * wrapper.audioFormat.mBytesPerFrame
        
        if wrapper.audioBufferList.mBuffers.mDataByteSize < expectedDataByteSize {
            wrapper.audioBufferList.mBuffers.mData = realloc(wrapper.audioBufferList.mBuffers.mData, Int(expectedDataByteSize))
            wrapper.audioBufferList.mBuffers.mDataByteSize = expectedDataByteSize
        }

        let status = wrapper.CheckErrorStatus(AudioUnitRender(wrapper.ioUnit!, ioActionFlags, inTimeStamp,
                                                          inBusNumber, inNumberFrames, &wrapper.audioBufferList),
                                          "AudioUnitRender调用")

        var audioDataFloat = [Float](repeating: 0.0, count: Int(inNumberFrames))
        let buffer = wrapper.audioBufferList.mBuffers
        let bufferData = buffer.mData!.assumingMemoryBound(to: Int16.self)
        
        for j in 0..<Int(buffer.mDataByteSize / UInt32(MemoryLayout<Int16>.size)) {
            audioDataFloat[j] = Float(bufferData[j]) / 32768.0 // 归一化到[-1.0, 1.0]范围
        }
        
        if status == noErr {
            wrapper.audioListQueue.async {
                wrapper.audioList.append(contentsOf: audioDataFloat)
            }
        }
        return status
    }

    private func CheckHasError(_ status: OSStatus, _ operation: String) -> Bool {
        if status != noErr {
            print("[AzureMicrophoneStream] \(operation)失败: \(status)")
            return true
        }
        return false
    }

    private func CheckErrorStatus(_ status: OSStatus, _ operation: String) -> OSStatus {
        if status != noErr {
            print("[AzureMicrophoneStream] \(operation)失败: \(status)")
        }
        return status
    }

    // 读取音频数据，适配Azure SDK
    func read(bytes: inout [UInt8]) -> Int {
        return audioListQueue.sync {
            if audioList.isEmpty {
                return 0
            }
            
            // 确保有足够的数据
            let minFrames = 1280
            if audioList.count < minFrames {
                return 0
            }
            
            let frameLength = minFrames
            
            let buffer = Array(audioList.prefix(frameLength))
            audioList.removeFirst(frameLength)
            
            // 转换为Int16数据
            var int16Data = buffer.map { Int16($0 * 32767) }
            let data = Data(buffer: UnsafeBufferPointer(start: &int16Data, count: int16Data.count))
            bytes = [UInt8](data)
            
            return frameLength * 2 // 每个样本2字节(16位PCM)
        }
    }
}

/// Azure ASR工具类，负责实现语音识别服务接口
@available(iOS 13.0, *)
class AzureAsrHelper: NSObject {
    // MARK: - 属性
    
    /// 事件处理回调
    private var eventHandler: (String, [String: Any]) -> Void
    
    /// 语音配置信息
    private var speechSubscriptionKey: String = ""
    private var serviceRegion: String = ""
    
    /// 语音识别相关
    private var speechConfig: SPXSpeechConfiguration?
    private var recognizer: SPXSpeechRecognizer?
    private var audioConfig: SPXAudioConfiguration?
    
    /// 麦克风流
    private var microphoneStream: AzureMicrophoneStream?
    private var pushStreamConfig: SPXPushAudioInputStream?
    
    /// 状态标志
    private var isInitialized = false
    private var _isContinuousRecognitionActive = false
    
    /// 当前语言和支持的语言
    private var currentLanguage = "zh-CN"
    private var supportedLanguages: [String] = ["zh-CN", "en-US"]
    private var isAutoDetectLanguage = false
    
    // MARK: - 初始化
    
    init(eventHandler: @escaping (String, [String: Any]) -> Void) {
        self.eventHandler = eventHandler
        super.init()
    }
    
    deinit {
        dispose()
    }
    
    // MARK: - ASR Service 接口实现
    
    /// 初始化语音识别服务
    /// - Parameters:
    ///   - speechSubscriptionKey: Azure 语音服务订阅密钥
    ///   - serviceRegion: Azure 服务区域 (如 eastasia)
    ///   - supportedLanguages: 支持的语言代码数组 (可选)
    /// - Returns: 初始化是否成功
    func initialize(speechSubscriptionKey: String, serviceRegion: String, supportedLanguages: [String]? = nil) -> Bool {
        print("[AzureAsrHelper] 初始化 Azure 语音服务")
        
        // 检查配置是否为空
        if speechSubscriptionKey.isEmpty || serviceRegion.isEmpty {
            print("[AzureAsrHelper] 错误: Azure 配置信息不完整")
            eventHandler("error", ["message": "Azure 配置信息不完整"])
            return false
        }
        
        // 释放之前的资源
        dispose()
        
        // 记录配置信息
        self.speechSubscriptionKey = speechSubscriptionKey
        self.serviceRegion = serviceRegion
        
        // 设置语言
        if let languages = supportedLanguages, !languages.isEmpty {
            self.supportedLanguages = languages
        }
        
        // 根据支持的语言数量决定是否启用自动语言检测
        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 {
            currentLanguage = self.supportedLanguages[0]
        }
        
        // 创建麦克风流
        microphoneStream = AzureMicrophoneStream()
        
        print("[AzureAsrHelper] Azure 语音服务初始化成功")
        isInitialized = true
        return true
    }
    
    /// 重置 recognizer
    private func resetRecognizer() -> Bool {
        // 释放之前的 recognizer
        recognizer = nil
        audioConfig = nil
        
        do {
            // 创建语音配置
            speechConfig = try SPXSpeechConfiguration(subscription: speechSubscriptionKey, region: serviceRegion)
            
            // 创建推送流
            pushStreamConfig = try SPXPushAudioInputStream()
            
            // 创建音频配置，使用推送流
            audioConfig = try SPXAudioConfiguration(streamInput: pushStreamConfig!)
            
            // 设置语言配置
            if isAutoDetectLanguage {
                // 设置自动语言检测
                speechConfig?.setPropertyTo("Continuous", by: SPXPropertyId.speechServiceConnectionLanguageIdMode)
                
                // 创建自动语言检测配置
                let autoDetectSourceLanguageConfig = try SPXAutoDetectSourceLanguageConfiguration(supportedLanguages)
                
                // 创建识别器
                recognizer = try SPXSpeechRecognizer(
                    speechConfiguration: speechConfig!, 
                    autoDetectSourceLanguageConfiguration: autoDetectSourceLanguageConfig,
                    audioConfiguration: audioConfig!
                )
            } else {
                // 设置指定的识别语言
                speechConfig?.speechRecognitionLanguage = currentLanguage
                
                // 创建识别器
                recognizer = try SPXSpeechRecognizer(speechConfiguration: speechConfig!, audioConfiguration: audioConfig!)
            }
            
            return true
        } catch {
            print("[AzureAsrHelper] 错误: 重置识别器失败: \(error.localizedDescription)")
            eventHandler("error", ["message": "重置识别器失败: \(error.localizedDescription)"])
            return false
        }
    }
    
    /// 启动音频捕获和数据推送
    private func startAudioStream() -> Bool {
        guard let micStream = microphoneStream else {
            print("[AzureAsrHelper] 错误: 麦克风流未初始化")
            return false
        }
        
        // 启动麦克风
        if !micStream.start() {
            print("[AzureAsrHelper] 错误: 启动麦克风流失败")
            return false
        }
        
        // 创建并启动音频推送线程
        DispatchQueue.global(qos: .userInitiated).async { [weak self] in
            guard let self = self, let pushStream = self.pushStreamConfig else { return }
            
            var isRunning = true
            var audioBuffer = [UInt8](repeating: 0, count: 16000)
            
            while isRunning {
                autoreleasepool {
                    // 读取麦克风数据
                    let bytesRead = micStream.read(bytes: &audioBuffer)
                    
                    if bytesRead > 0 {
                        do {
                            // 推送音频数据到Azure识别流
                            let data = Data(bytes: audioBuffer, count: bytesRead)
                            try pushStream.write(data)
                        } catch {
                            print("[AzureAsrHelper] 推送音频数据失败: \(error.localizedDescription)")
                            isRunning = false
                        }
                    }
                    
                    // 检查是否应该继续捕获
                    if !self._isContinuousRecognitionActive {
                        isRunning = false
                    }
                    
                    // 添加适当的休眠以避免过度消耗CPU
                    if bytesRead == 0 {
                        Thread.sleep(forTimeInterval: 0.01)
                    }
                }
            }
            print("[AzureAsrHelper] 音频推送线程已停止")
        }
        
        return true
    }
    
    /// 执行一次性语音识别
    /// - Returns: 是否成功启动识别
    func recognizeOnce() -> Bool {
        if !isInitialized {
            print("[AzureAsrHelper] 错误: 语音服务未初始化")
            eventHandler("error", ["message": "语音服务未初始化"])
            return false
        }
        
        // 如果正在连续识别，先停止
        if _isContinuousRecognitionActive {
            stopContinuousRecognition()
        }
        
        // 重置 recognizer
        if !resetRecognizer() {
            return false
        }
        
        // 启动音频流
        if !startAudioStream() {
            return false
        }
        
        do {
            // 设置回调
            recognizer?.addRecognizedEventHandler { [weak self] _, event in
                guard let self = self else { return }
                
                if event.result.reason == SPXResultReason.recognizedSpeech {
                    let detectedLanguage = self.getDetectedLanguage(from: event.result)
                    self.eventHandler("result", [
                        "text": event.result.text ?? "",
                        "detectedLanguage": detectedLanguage
                    ])
                }
            }
            
            recognizer?.addCanceledEventHandler { [weak self] _, event in
                guard let self = self else { return }
                
                let errorDetails = event.errorDetails ?? "未知错误"
                self.eventHandler("error", ["message": "识别异常: \(errorDetails)"])
            }
            
            // 通知会话开始
            eventHandler("sessionStarted", [:])
            
            // 执行识别
            try recognizer?.recognizeOnceAsync { [weak self] result in
                guard let self = self else { return }
                
                // 停止麦克风流
                self.microphoneStream?.stop()
                
                if result.reason == SPXResultReason.recognizedSpeech {
                    let detectedLanguage = self.getDetectedLanguage(from: result)
                    self.eventHandler("result", [
                        "text": result.text ?? "",
                        "detectedLanguage": detectedLanguage
                    ])
                } else if result.reason == SPXResultReason.canceled {
                    do {
                        let details = try SPXCancellationDetails(fromCanceledRecognitionResult: result)
                        let errorDetails = details.errorDetails ?? "未知错误"
                        self.eventHandler("error", ["message": "识别取消: \(errorDetails)"])
                    } catch {
                        print("[AzureAsrHelper] 错误: 获取取消详情失败: \(error.localizedDescription)")
                        self.eventHandler("error", ["message": "识别取消，无法获取详细原因"])
                    }
                }
            }
            
            return true
        } catch {
            print("[AzureAsrHelper] 错误: 识别异常: \(error.localizedDescription)")
            eventHandler("error", ["message": "识别异常: \(error.localizedDescription)"])
            microphoneStream?.stop()
            return false
        }
    }
    
    /// 开始连续语音识别
    /// - Returns: 是否成功启动识别
    func startContinuousRecognition() -> Bool {
        if !isInitialized {
            print("[AzureAsrHelper] 错误: 语音服务未初始化")
            eventHandler("error", ["message": "语音服务未初始化"])
            return false
        }
        
        // 如果已经在进行连续识别，先停止
        if _isContinuousRecognitionActive {
            stopContinuousRecognition()
        }
        
        // 重置 recognizer
        if !resetRecognizer() {
            return false
        }
        
        do {
            // 设置识别事件处理
            setupContinuousRecognitionCallbacks()
            
            // 启动连续识别
            try recognizer?.startContinuousRecognition()
            _isContinuousRecognitionActive = true
            
            // 启动音频流
            if !startAudioStream() {
                try recognizer?.stopContinuousRecognition()
                _isContinuousRecognitionActive = false
                return false
            }
            
            // 通知会话开始
            eventHandler("sessionStarted", [:])
            
            print("[AzureAsrHelper] 连续识别开始")
            return true
        } catch {
            print("[AzureAsrHelper] 错误: 开始连续识别失败: \(error.localizedDescription)")
            eventHandler("error", ["message": "开始连续识别失败: \(error.localizedDescription)"])
            _isContinuousRecognitionActive = false
            microphoneStream?.stop()
            return false
        }
    }
    
    /// 停止连续语音识别
    /// - Returns: 是否成功停止识别
    func stopContinuousRecognition() -> Bool {
        if !_isContinuousRecognitionActive || recognizer == nil {
            return true
        }
        
        // 停止麦克风流
        microphoneStream?.stop()
        
        do {
            try recognizer?.stopContinuousRecognition()
            _isContinuousRecognitionActive = false
            eventHandler("sessionStopped", [:])
            print("[AzureAsrHelper] 连续识别已停止")
            return true
        } catch {
            print("[AzureAsrHelper] 错误: 停止连续识别失败: \(error.localizedDescription)")
            eventHandler("error", ["message": "停止连续识别失败: \(error.localizedDescription)"])
            _isContinuousRecognitionActive = false
            return false
        }
    }
    
    /// 检查连续识别是否活跃
    /// - Returns: 连续识别是否处于活跃状态
    func isContinuousRecognitionActive() -> Bool {
        return _isContinuousRecognitionActive
    }
    
    /// 释放资源
    func dispose() {
        print("[AzureAsrHelper] 释放资源")
        
        // 停止连续识别
        if _isContinuousRecognitionActive {
            stopContinuousRecognition()
        }
        
        // 关闭麦克风流
        microphoneStream?.dispose()
        microphoneStream = nil
        
        // 关闭推送流
        if let pushStream = pushStreamConfig {
            do {
                try pushStream.close()
            } catch {
                print("[AzureAsrHelper] 关闭推送流失败: \(error.localizedDescription)")
            }
        }
        
        // 释放资源
        recognizer = nil
        speechConfig = nil
        audioConfig = nil
        pushStreamConfig = nil
        
        // 重置状态
        _isContinuousRecognitionActive = false
        isInitialized = false
    }
    
    // MARK: - 私有辅助方法
    
    /// 设置连续识别回调
    private func setupContinuousRecognitionCallbacks() {
        // 最终识别结果
        recognizer?.addRecognizedEventHandler { [weak self] _, event in
            guard let self = self else { return }
            
            if event.result.reason == SPXResultReason.recognizedSpeech {
                let detectedLanguage = self.getDetectedLanguage(from: event.result)
                print("[AzureAsrHelper] 识别结果: \(event.result.text ?? ""), 语言: \(detectedLanguage)")
                self.eventHandler("result", [
                    "text": event.result.text ?? "",
                    "detectedLanguage": detectedLanguage
                ])
            }
        }
        
        // 识别中事件
        recognizer?.addRecognizingEventHandler { [weak self] _, event in
            guard let self = self else { return }
            
            if event.result.reason == SPXResultReason.recognizingSpeech {
                let detectedLanguage = self.getDetectedLanguage(from: event.result)
                print("[AzureAsrHelper] 识别中: \(event.result.text ?? ""), 语言: \(detectedLanguage)")
                self.eventHandler("recognizing", [
                    "text": event.result.text ?? "",
                    "detectedLanguage": detectedLanguage
                ])
            }
        }
        
        // 会话事件
        recognizer?.addSessionStartedEventHandler { [weak self] _, _ in
            guard let self = self else { return }
            
            self._isContinuousRecognitionActive = true
            self.eventHandler("sessionStarted", [:])
        }
        
        recognizer?.addSessionStoppedEventHandler { [weak self] _, _ in
            guard let self = self else { return }
            
            self._isContinuousRecognitionActive = false
            self.eventHandler("sessionStopped", [:])
        }
        
        // 取消事件
        recognizer?.addCanceledEventHandler { [weak self] _, event in
            guard let self = self else { return }
            
            let reason = event.reason.rawValue
            let errorDetails = event.errorDetails ?? ""
            
            self.eventHandler("canceled", [
                "reason": reason,
                "errorDetails": errorDetails
            ])
            
            self._isContinuousRecognitionActive = false
            self.microphoneStream?.stop()
        }
    }
    
    /// 从结果中获取检测到的语言
    private func getDetectedLanguage(from result: SPXSpeechRecognitionResult) -> String {
        if isAutoDetectLanguage {
            do {
                let langResult = try SPXAutoDetectSourceLanguageResult(result)
                return langResult.language ?? currentLanguage
            } catch {
                print("[AzureAsrHelper] 错误: 获取检测到的语言失败: \(error.localizedDescription)")
                return currentLanguage
            }
        } else {
            return currentLanguage
        }
    }
}