We are running a custom speech model use Azure's Cognitive Services / Speech-to-text and I'm running across an issue where users reported "proceed" is transcribed as "pro ceed". As it turns out, our related text has "pro ceed" in it - which explains the cause.
However, the related text was originally uploaded with the term "proceed" (not "pro ceed"). Here is a sample with just the single word:
The text file is UTF-8 with BOM
Has anyone else seen this?