Speech-to-Text — Gotchas & Best Practices

Speech-to-Text — Gotchas & Best Practices

For full service implementation, types, and UI integration patterns, see references/​speech-to-text-api.md.

Critical: Permission Handling

Always request permissions before starting speech recognition. Both microphone and speech permissions are required.

// ❌ Starting recognition without checking permissions
var result = await _speechService.StartListeningAsync();

// ✅ Always check permissions first
if (!await _speechService.RequestPermissionsAsync())
    return; // Gracefully handle denial
var result = await _speechService.StartListeningAsync(cancellationToken);

⚠️ iOS Requires Both Permission Descriptions

Missing either NSSpeechRecognitionUsageDescription or NSMicrophoneUsageDescription in Info.plist will cause a runtime crash — not a graceful failure.

⚠️ Android: Permission Re-prompt

On Android, if the user denies the RECORD_AUDIO permission twice, the OS stops showing the prompt. You must guide users to Settings manually.

Common Mistakes

❌ No Timeout — Indefinite Listening

Always set a timeout to prevent indefinite listening sessions that drain battery:

// ❌ No timeout — listens forever if no speech detected
await _speechToText.StartListenAsync(options, CancellationToken.None);

// ✅ Use a combined timeout + user cancellation token
using var timeoutCts = new CancellationTokenSource(TimeSpan.FromSeconds(60));
using var combinedCts = CancellationTokenSource.CreateLinkedTokenSource(
    userCancellationToken, timeoutCts.Token);
await _speechToText.StartListenAsync(options, combinedCts.Token);

❌ Not Unsubscribing Event Handlers

Leaking event subscriptions causes duplicate processing and memory leaks:

// ❌ Subscribe without unsubscribe
_speechToText.RecognitionResultUpdated += OnRecognitionResultUpdated;

// ✅ Always unsubscribe in finally block
try
{
    _speechToText.RecognitionResultUpdated += OnRecognitionResultUpdated;
    // ... listen ...
}
finally
{
    _speechToText.RecognitionResultUpdated -= OnRecognitionResultUpdated;
}

❌ Not Disposing CancellationTokenSource

// ❌ Leaked CTS
_currentCts = new CancellationTokenSource();

// ✅ Dispose in finally
try { /​* ... */ }
finally
{
    _currentCts?.Dispose();
    _currentCts = null;
}

Platform Pitfalls

Architecture Tips

1. Wrap ISpeechToText in a service — Don't use SpeechToText.Default directly in ViewModels. Wrap in ISpeechRecognitionService for testability and state management.

2. Use partial results for UX — Subscribe to PartialResultReceived for live transcription feedback. Users expect to see words appear as they speak.

3. Continuous listening = loop with delay — Loop StartListeningAsync with small delays (Task.Delay(100)) for conversation mode.

4. Guard against double-start — Check state before starting:

   if (State == SpeechRecognitionState.Listening)
       return new SpeechRecognitionResultDto { Success = false, ErrorMessage = "Already listening" };
   

5. Natural language output — CommunityToolkit.Maui returns normalized, punctuated text — not raw phonemes. No post-processing needed for basic use cases.

6. UI-agnostic service — The ISpeechRecognitionService pattern works identically with XAML/​MVVM, C# Markup, and MauiReactor. See references/​speech-to-text-api.md for all three patterns.

Checklist

• CommunityToolkit.Maui NuGet installed (look up current version)
• UseMauiCommunityToolkit() called in MauiProgram.cs
• ISpeechToText registered as singleton via DI
• iOS: Both NSSpeechRecognitionUsageDescription and NSMicrophoneUsageDescription in Info.plist
• Android: RECORD_AUDIO permission in AndroidManifest.xml
• Permissions checked before every StartListeningAsync call
• Timeout configured (recommend 60 seconds max)
• Event handlers unsubscribed in finally blocks
• CancellationTokenSource disposed after use