SFTranscriptionSegment confidence is always 0 for Saudi Arabia, but ok for USA

Hello,

I have SFSpeechRecognizer initialised with Locale(identifier: "en-US")

After setting Region (on phone settings) to one of these countries:

• Saudi Arabia

• United Arab Emirates

• Malaysia

SFTranscriptionSegment always have confidence 0.

Setting Region back to United States (or Russia) makes confidence correct (non-zero value).

This issue is 100% reproducible on Simulator & Device with SpeakToMe sample App by Apple https://developer.apple.com/library/archive/samplecode/SpeakToMe/Introduction/Intro.html

I'm adopted it to Swift 5 and added logging. Modified Sample project's ViewController.swift is attached.


To reproduce:

1. Tap "Start recording"

2. Say "My name is Max"

3. Tap "Stop recording"

Last output for Saudi Arabia:

New result from recognition task
Segment confidence: 0.0
Segment confidence: 0.0
Segment confidence: 0.0
Segment confidence: 0.0
Segment confidence: 0.0

Last output for United States:

New result from recognition task
Segment confidence: 0.925
Segment confidence: 0.929
Segment confidence: 0.929
Segment confidence: 0.931
Segment confidence: 0.922