[BUG] SpeechToText on iOS 17 cuts off or fails to recognize many words

[AlessandroLanzoni1991]

Is there an existing issue for this?

• I have searched the existing issues

Did you read the "Reporting a bug" section on Contributing file?

• I have read the "Reporting a bug" section on Contributing file: https://github.com/CommunityToolkit/Maui/blob/main/CONTRIBUTING.md#reporting-a-bug

Current Behavior

I am currently experiencing an issue with SpeechToText on iOS 17 where during voice recognition, certain words are either cut off or not recognized correctly.
For example:
"Alessandro" become Al or Alex.
"Planification" become "plan".
"Configuration" become "config"

Expected Behavior

All spoken words should be transcribed accurately without being cut off or distorted during speech recognition.

Steps To Reproduce

1. Open any application that uses SpeechToText functionality on an iOS 17 device.
2. Speak various sentences and words.
3. Notice how many words are truncated or inaccurately transcribed during the speech recognition process, for example Names

Link to public reproduction project repository

https://github.com/jfversluis/MauiSpeechToTextSample

Environment

- .NET MAUI CommunityToolkit: 9.0.1
- OS: 17.5.1    / Iphone 12
- .NET MAUI: 8.0.60

Anything else?

This works well on my iPhone 8 whit iOS 16.7.6

[vhugogarcia]

Thanks @AlessandroLanzoni1991 for your feedback. I was able to verify the issue. However, I noticed that it is an expected result as it is right now configured.

When creating a new SFSpeechAudioBufferRecognitionRequest(); in the code below, Apple by default sets the ShouldReportPartialResults to True. I tested it by running the sample project of the MCT and setting that property to false and it worked OK. The only downtime of that is that it has a delay which prevents to show the text in real time.

By saying that, I think it will be ideal to add a property where we can set that to True or False based on the user's requirements.

Maui/src/CommunityToolkit.Maui.Core/Essentials/SpeechToText/SpeechToTextImplementation.ios.cs

Line 24 in 819ab10

liveSpeechRequest = new SFSpeechAudioBufferRecognitionRequest();

Thanks

[Spoxeman]

I am experiencing the same issues, Is there any resolution yet?

@vhugogarcia ..I cloned the ios implementation and made the changes to set ShouldReportPartialResults to false and it works, no more truncation. How do we get this into release?

[VladislavAntonyuk]

Working on it

[aaldrichaes2]

Is there a workaround for this currently?

[VladislavAntonyuk]

@aaldrichaes2 It is just an intermediate result and the final recognition text will differ and should be correct.