How to convert speech to text using SFSpeechRecognizer

Swift version: 5.6

Paul Hudson @twostraws May 28th 2019

iOS has a built-in speech transcription system, which allows you to convert any audio recording into a text stream. It takes a few steps to configure, so let’s walk through them.

First, add import Speech to the top of your Swift file, to bring in the Speech framework.

Second, request permission to transcribe audio:

func requestTranscribePermissions() {
    SFSpeechRecognizer.requestAuthorization { [unowned self] authStatus in
        DispatchQueue.main.async {
            if authStatus == .authorized {
                print("Good to go!")
            } else {
                print("Transcription permission was declined.")
            }
        }
    }
}

Third, add a key to your Info.plist called NSSpeechRecognitionUsageDescription, then give it a string describing what you intend to do with the transcriptions.

Finally, write a method to perform transcription on an audio URL. This URL should be a recording you’ve already made, that is stored locally on the device:

func transcribeAudio(url: URL) {
    // create a new recognizer and point it at our audio
    let recognizer = SFSpeechRecognizer()
    let request = SFSpeechURLRecognitionRequest(url: url)

    // start recognition!
    recognizer?.recognitionTask(with: request) { [unowned self] (result, error) in
        // abort if we didn't get any transcription back
        guard let result = result else {
            print("There was an error: \(error!)")
            return
        }

        // if we got the final transcription back, print it
        if result.isFinal {
            // pull out the best transcription...
            print(result.bestTranscription.formattedString)
        }
    }
}

Note: the isFinal property is there because you may get an initial transcription back containing some or all of the text, but it’s only considered final – i.e. as good as it gets – when the isFinal flag is true.