iOS Speech框架实战：语音转文字的完整实现指南

2025年11月14日 互联网

iOS Speech框架实战：语音转文字的完整实现指南

一、Speech框架核心价值与适用场景

Speech框架是Apple在iOS 10中引入的语音识别API，相较于传统第三方SDK，其优势体现在三个方面：

  1. 系统级集成：直接调用设备内置语音引擎，无需网络请求（离线模式）
  2. 隐私保护：所有语音处理在本地完成，符合App Store隐私政策要求
  3. 性能优化：针对A系列芯片深度优化，延迟控制在200ms以内

典型应用场景包括：

  • 实时会议记录（如配合ReplayKit实现屏幕共享+语音转写）
  • 无障碍功能开发（为视障用户提供语音导航）
  • 健身应用指令识别（如”开始跑步”、”暂停计时”）
  • 医疗行业术语转录（需配合自定义词汇表）

二、基础环境配置

1. 权限声明

在Info.plist中添加两个必要键值：

  1. <key>NSSpeechRecognitionUsageDescription</key>
  2. <string>需要麦克风权限实现语音转文字功能</string>
  3. <key>NSMicrophoneUsageDescription</key>
  4. <string>需要访问麦克风以捕获语音输入</string>

2. 框架导入

  1. import Speech

3. 权限请求流程

  1. func requestSpeechRecognitionPermission() {
  2.     SFSpeechRecognizer.requestAuthorization { authStatus in
  3.         DispatchQueue.main.async {
  4.             switch authStatus {
  5.             case .authorized:
  6.                 print("语音识别权限已授权")
  7.             case .denied:
  8.                 print("用户拒绝权限")
  9.             case .restricted:
  10.                 print("设备限制语音识别")
  11.             case .notDetermined:
  12.                 print("权限状态未确定")
  13.             @unknown default:
  14.                 break
  15.             }
  16.         }
  17.     }
  18. }

三、核心功能实现

1. 基础语音转写

  1. let audioEngine = AVAudioEngine()
  2. let speechRecognizer = SFSpeechRecognizer(locale: Locale(identifier: "zh-CN"))
  3. var recognitionRequest: SFSpeechAudioBufferRecognitionRequest?
  4. var recognitionTask: SFSpeechRecognitionTask?
  6. func startRecording() {
  7.     // 配置音频会话
  8.     let audioSession = AVAudioSession.sharedInstance()
  9.     try! audioSession.setCategory(.record, mode: .measurement, options: .duckOthers)
  10.     try! audioSession.setActive(true, options: .notifyOthersOnDeactivation)
  12.     // 创建识别请求
  13.     recognitionRequest = SFSpeechAudioBufferRecognitionRequest()
  14.     guard let recognitionRequest = recognitionRequest else { return }
  16.     // 设置识别任务
  17.     recognitionTask = speechRecognizer?.recognitionTask(with: recognitionRequest) { result, error in
  18.         if let result = result {
  19.             let transcribedText = result.bestTranscription.formattedString
  20.             print("转写结果: \(transcribedText)")
  22.             // 实时更新UI需在主线程
  23.             DispatchQueue.main.async {
  24.                 self.textView.text = transcribedText
  25.             }
  26.         }
  28.         if error != nil {
  29.             self.stopRecording()
  30.             print("识别错误: \(error?.localizedDescription ?? "")")
  31.         }
  32.     }
  34.     // 配置音频引擎
  35.     let inputNode = audioEngine.inputNode
  36.     let recordingFormat = inputNode.outputFormat(forBus: 0)
  37.     inputNode.installTap(onBus: 0, bufferSize: 1024, format: recordingFormat) { buffer, _ in
  38.         recognitionRequest.append(buffer)
  39.     }
  41.     audioEngine.prepare()
  42.     try! audioEngine.start()
  43. }
  45. func stopRecording() {
  46.     audioEngine.stop()
  47.     recognitionRequest?.endAudio()
  48.     recognitionTask?.finish()
  49.     recognitionTask = nil
  50.     recognitionRequest = nil
  51. }

2. 高级功能实现

多语言支持

  1. // 支持中英文混合识别
  2. let mixedLocale = Locale(identifier: "zh-Hans-CN") // 中文为主
  3. let speechRecognizer = SFSpeechRecognizer(locale: mixedLocale)
  5. // 或动态切换语言
  6. func switchLanguage(to localeIdentifier: String) {
  7.     guard let newRecognizer = SFSpeechRecognizer(locale: Locale(identifier: localeIdentifier)) else {
  8.         print("不支持该语言")
  9.         return
  10.     }
  11.     speechRecognizer = newRecognizer
  12.     // 需重新创建recognitionTask
  13. }

自定义词汇表

  1. // 创建词汇表
  2. let vocabulary = SFSpeechRecognitionVocabulary()
  3. vocabulary.addTerm("Xcode")  // 添加专业术语
  4. vocabulary.addTerm("SwiftUI")
  6. // 应用到识别器
  7. let config = SFSpeechRecognizer.supportedLocales().first { $0.identifier == "zh-CN" }
  8. let customRecognizer = SFSpeechRecognizer(locale: config!)
  9. customRecognizer?.supportsOnDeviceRecognition = true  // 启用离线模式

实时处理优化

  1. // 使用分段回调提高响应速度
  2. recognitionTask = speechRecognizer?.recognitionTask(with: recognitionRequest, 
  3.     delegate: self)  // 需实现SFSpeechRecognitionTaskDelegate
  5. extension ViewController: SFSpeechRecognitionTaskDelegate {
  6.     func speechRecognitionTask(_ task: SFSpeechRecognitionTask, 
  7.         didHypothesizeTranscription transcription: SFTranscription) {
  8.         // 收到临时识别结果时调用
  9.         let partialText = transcription.formattedString
  10.         updateUI(with: partialText)
  11.     }
  13.     func speechRecognitionTaskFinishedReadingAudio(_ task: SFSpeechRecognitionTask) {
  14.         print("音频输入完成")
  15.     }
  16. }

四、错误处理与性能优化

1. 常见错误处理

错误类型 处理方案
SFSpeechRecognizerError.notReady 检查麦克风权限和网络连接
SFSpeechRecognizerError.audioBufferTooSmall 调整bufferSize参数（建议512-2048）
SFSpeechRecognizerError.requestTimeout 设置recognitionRequest.shouldReportPartialResults = true
音频引擎启动失败 检查AVAudioSession配置，确保无其他应用占用麦克风

2. 性能优化技巧

  1. 离线模式配置：

    1. if let onDeviceRecognizer = SFSpeechRecognizer(locale: Locale(identifier: "zh-CN")) {
    2.  onDeviceRecognizer.supportsOnDeviceRecognition = true
    3.  // 优先使用离线识别
    4. }

  2. 内存管理：

  • 及时调用finish()方法释放识别任务
  • 使用弱引用避免循环引用 
    1. weak var weakSelf = self
    2. recognitionTask?.finish()
    3. weakSelf?.recognitionTask = nil
  1. 电池优化：
  • 在后台任务中暂停识别
  • 监听设备电量变化动态调整识别精度

五、完整实现示例

  1. import UIKit
  2. import Speech
  3. import AVFoundation
  5. class SpeechRecognitionViewController: UIViewController {
  7.     @IBOutlet weak var textView: UITextView!
  8.     @IBOutlet weak var recordButton: UIButton!
  10.     private let audioEngine = AVAudioEngine()
  11.     private var speechRecognizer: SFSpeechRecognizer?
  12.     private var recognitionRequest: SFSpeechAudioBufferRecognitionRequest?
  13.     private var recognitionTask: SFSpeechRecognitionTask?
  14.     private var isRecording = false
  16.     override func viewDidLoad() {
  17.         super.viewDidLoad()
  18.         setupSpeechRecognizer()
  19.         requestSpeechRecognitionPermission()
  20.     }
  22.     private func setupSpeechRecognizer() {
  23.         let locale = Locale(identifier: "zh-CN")
  24.         speechRecognizer = SFSpeechRecognizer(locale: locale)
  25.     }
  27.     private func requestSpeechRecognitionPermission() {
  28.         SFSpeechRecognizer.requestAuthorization { authStatus in
  29.             DispatchQueue.main.async {
  30.                 guard authStatus == .authorized else {
  31.                     self.showPermissionAlert()
  32.                     return
  33.                 }
  34.                 self.recordButton.isEnabled = true
  35.             }
  36.         }
  37.     }
  39.     private func showPermissionAlert() {
  40.         let alert = UIAlertController(title: "权限不足", 
  41.             message: "请在设置中开启麦克风权限", 
  42.             preferredStyle: .alert)
  43.         alert.addAction(UIAlertAction(title: "确定", style: .default))
  44.         present(alert, animated: true)
  45.     }
  47.     @IBAction func toggleRecording(_ sender: UIButton) {
  48.         isRecording.toggle()
  50.         if isRecording {
  51.             startRecording()
  52.             sender.setTitle("停止", for: .normal)
  53.         } else {
  54.             stopRecording()
  55.             sender.setTitle("开始", for: .normal)
  56.         }
  57.     }
  59.     private func startRecording() {
  60.         // 配置音频会话
  61.         let audioSession = AVAudioSession.sharedInstance()
  62.         try! audioSession.setCategory(.record, mode: .measurement, options: .duckOthers)
  63.         try! audioSession.setActive(true)
  65.         // 创建识别请求
  66.         recognitionRequest = SFSpeechAudioBufferRecognitionRequest()
  67.         guard let recognitionRequest = recognitionRequest else { return }
  69.         // 设置识别任务
  70.         recognitionTask = speechRecognizer?.recognitionTask(with: recognitionRequest) { result, error in
  71.             if let result = result {
  72.                 let text = result.bestTranscription.formattedString
  73.                 DispatchQueue.main.async {
  74.                     self.textView.text = text
  75.                 }
  76.             }
  78.             if error != nil {
  79.                 self.stopRecording()
  80.                 print("识别错误: \(error?.localizedDescription ?? "")")
  81.             }
  82.         }
  84.         // 配置音频引擎
  85.         let inputNode = audioEngine.inputNode
  86.         let recordingFormat = inputNode.outputFormat(forBus: 0)
  87.         inputNode.installTap(onBus: 0, bufferSize: 1024, format: recordingFormat) { buffer, _ in
  88.             recognitionRequest.append(buffer)
  89.         }
  91.         audioEngine.prepare()
  92.         try! audioEngine.start()
  93.     }
  95.     private func stopRecording() {
  96.         audioEngine.stop()
  97.         recognitionRequest?.endAudio()
  98.         recognitionTask?.finish()
  99.         recognitionTask = nil
  100.         recognitionRequest = nil
  101.     }
  103.     deinit {
  104.         stopRecording()
  105.         try? AVAudioSession.sharedInstance().setActive(false)
  106.     }
  107. }

六、最佳实践建议

  1. 离线优先策略：在支持的设备上优先使用supportsOnDeviceRecognition
  2. 动态语言检测：通过SFTranscription的segment属性分析语言变化
  3. 省电模式：当设备电量低于20%时自动降低识别精度
  4. 用户引导：首次使用时展示麦克风权限获取的明确说明
  5. 结果校验：对专业术语使用正则表达式进行二次校验

通过Speech框架实现的语音转文字功能，在保持高准确率的同时，能有效保护用户隐私。开发者应根据具体场景选择在线/离线模式，并合理处理权限请求流程，以提供流畅的用户体验。

最新文章
热门标签