MAUI 语音识别 - 启用 SpeechInputCompleteSilence 意图时无法存储已处理语音的最终结果

Question

我正在使用 Gerald Versluis 方法在 Android 上使用 MAUI 启用语音识别和语音转文本 - https://github.com/jfversluis/MauiSpeechToTextSample

我试图通过将识别文本移动到另一个标签来使其永久化，这样它就不会被新传入的语音覆盖，但我似乎无法访问保存最终处理语音的值。

在RecordButton方法中，包含最终识别语音的值是RecognitionText。来自 MainPage.xaml.cs

    private async void RecordButton()
    {
        var isAuthorized = await speechToText.RequestPermissions();

        if (isAuthorized)
        {
            try
            {
                RecognitionText = await speechToText.Listen(
                    System.Globalization.CultureInfo.GetCultureInfo("en-us"),
                    new Progress<string>(partialText =>
                    {
                        RecognitionText = partialText;
                        OnPropertyChanged(nameof(RecognitionText));
                    }),
                    tokenSource.Token);

// When ExtraSpeechInputCompleteSilenceLengthMillis is used, this code is never reached
// When ExtraSpeechInputCompleteSilenceLengthMillis is commented out, this code is reached and everything works great
                HistoryText += RecognitionText + Environment.NewLine + Environment.NewLine;
                OnPropertyChanged(nameof(HistoryText));
            }
            catch (Exception ex)
            {
                await DisplayAlert("Error", ex.ToString(), "OK");
            }
        }
        else
        {
            await DisplayAlert("Permission Error", "No microphone access", "OK");
        }
    }

这是 MainPage.xaml 中的 XAML 部分

    <Grid>
        <Grid.RowDefinitions>
            <RowDefinition Height="3*" />
            <RowDefinition Height="*" />
            <RowDefinition Height="*" />
        </Grid.RowDefinitions>
        <Grid.ColumnDefinitions>
            <ColumnDefinition Width="*" />
        </Grid.ColumnDefinitions>

        <Border>
            <ScrollView Margin="10">
                <Label
                    Grid.Row="0"
                    x:Name="HistoryLabel"
                    Text="{Binding HistoryText}"
                    FontSize="{OnIdiom Phone=Medium, Tablet=Large}" />
            </ScrollView>
        </Border>
        <Label
            Grid.Row="1"
            x:Name="RecognitionLabel"
            Text="{Binding RecognitionText}"
            FontSize="{OnIdiom Phone=Medium, Tablet=Large}"
            Margin="5"
            Padding="5" />
        <Button
            Grid.Row="2"
            CornerRadius="60"
            WidthRequest="120"
            HeightRequest="120"
            Text="Record"
            FontSize="24"
            FontAttributes="Bold"
            Command="{Binding RecordButtonCommand}" />

    </Grid>

这是 SpeechToTextImplementation.cs 中的 Listen 方法定义

        public async Task<string> Listen(CultureInfo culture, IProgress<string> recognitionResult, CancellationToken cancellationToken)
        {
            var taskResult = new TaskCompletionSource<string>();

            listener = new SpeechRecognitionListener
            {
                Error = ex => taskResult.TrySetException(new Exception("Failure in speech engine - " + ex)),
                PartialResults = sentence =>
                {
                    recognitionResult?.Report(sentence);
                },
                Results = sentence => taskResult.TrySetResult(sentence)
            };

            speechRecognizer = SpeechRecognizer.CreateSpeechRecognizer(Android.App.Application.Context);

            if (speechRecognizer is null)
                throw new ArgumentException("Speech recognizer is not available.");

            speechRecognizer.SetRecognitionListener(listener);
            speechRecognizer.StartListening(CreateSpeechIntent(culture));

            await using (cancellationToken.Register(() =>
            {
                StopRecording();
                taskResult.TrySetCanceled();
            }))
            {
                return await taskResult.Task;
            }
        }

最后是具有问题意图的 CreateSpeechIntent 方法

        private static Intent CreateSpeechIntent(CultureInfo culture)
        {
            var intent = new Intent(RecognizerIntent.ActionRecognizeSpeech);
            intent.PutExtra(RecognizerIntent.ExtraLanguagePreference, Java.Util.Locale.Default);

            var javaLocale = Java.Util.Locale.ForLanguageTag(culture.Name);
            intent.PutExtra(RecognizerIntent.ExtraLanguage, javaLocale);
            intent.PutExtra(RecognizerIntent.ExtraLanguageModel, RecognizerIntent.LanguageModelFreeForm);
            intent.PutExtra(RecognizerIntent.ExtraCallingPackage, Android.App.Application.Context.PackageName);

//-->Problem//intent.PutExtra(RecognizerIntent.ExtraSpeechInputCompleteSilenceLengthMillis, 10000);

            //intent.PutExtra(RecognizerIntent.ExtraBiasingStrings, ...) // SCA Use ATC strings!!
            //intent.PutExtra(RecognizerIntent.ExtraMaxResults, 1);
            //intent.PutExtra(RecognizerIntent.ExtraSpeechInputPossiblyCompleteSilenceLengthMillis, 1500);
            //intent.PutExtra(RecognizerIntent.ExtraSpeechInputMinimumLengthMillis, 15000); // SCA Are you done talking? Yes or no. Set right ere.

            intent.PutExtra(RecognizerIntent.ExtraPartialResults, true);

            return intent;
        }

• 注释掉intent.PutExtra(RecognizerIntent.ExtraSpeechInputCompleteSilenceLengthMillis, 10000);解决问题。不幸的是，我需要让麦克风打开 10 秒钟。
• 我对 ExtraSpeechInputCompleteSilenceLengthMillis 使用了不同的值，但我使用的任何值都会导致问题。这仅用于调试，因为我需要整整 10 秒。
• 我尝试从 SpeecToTextImplementation.cs 访问 RecognitionText，但无论我尝试什么，UI 都不会使用发送的值进行更新。 SpeecToTextImplementation.cs 位于 Android 文件夹中。

Answer 1

只需使用 MAUI 社区工具包即可轻松解决此问题。 MCT 拥有处理语音处理所需的所有方法。不再需要手动执行此操作。只需使用工具包事件处理程序 SpeechToTextRecognitionResultUpdatedEventArgs。