The average accuracy of speech recognition is about 90%, which may require manual corrections. Accuracy depends, among other things, on the speakers' speaking ability, ambient noise, quality of microphones, quality of speech regeneration, quality of info-communication lines/channels, etc.
Online speech recognition and translation is provided through Microsoft or Google's systems on servers abroad, for which you can choose from our monthly subscriptions in our Related Products or register directly to Microsoft Azure / Google Cloud.
Basic speech recognition without a language model of your own (language training).