Generate Japanese transcription with related text files when using Custom Speech
Currently, after using Katakana in the related text files to train a Custom Speech model, if generate transcription from audio files with this custom model, some Katakana in the related text files won't be generated as expected.
Use Katakana below to train a custom model.
Then, if we use the trained model to generate transcription from an audio file which contains "ソラ", "空" will be generated, instead of "ソラ".
But in some use cases, "ソラ" maybe a brand name or anything else technical terms. We expect to get “ソラ” with the trained custom model.
So, please consider improving this to generate transcription with the related text files.