speech batch transcription submit one by one or in batches.

That really depends on your application design (how fast you want users to be able to see results), and if you want to have your service running all the time or on a schedule X times a day. one other alternative to the REST batch service, especially if you have multiple people speaking in the audio, we have a batch api available in the speech sdk for conversation transcription, and it is a little simpler to use that the rest api. If that sounds like it would work, take a look at this example. https://github.com/Azure-Samples/cognitive-services-speech-sdk/blob/master/samples/csharp/sharedcontent/console/remote_conversation_samples.cs