We noticed in our apps that from iOS 13 on we only can use the (German) speech recognition API every 30 seconds or so. After the first run, every new request just fails. Only if we wait for approx 30 seconds speech is transcribed again. We do not have this problem with iOS 12.
With the demo application on https://developer.apple.com/documentation/speech/recognizing_speech_in_live_audio you can test this. On iOS 12 you can immediately start a new transcription after stopping the recording. On iOS 13 you have to wait 30 seconds after each recording.
This is a huge inconvenience for our users. Is there a bugfix on the way?