ConformerCTC for streaming

Is there a way to train a Conformer model with CTC loss function, such that when inferring live using blocked buffered data, you get the same output as if passing the whole data in one go. Also, could this be resilient to sample offsets?

I would like to use a Conformer model trained with CTC loss live using buffered data coming of a sensor.

There are a few papers on this already, such as https://arxiv.org/pdf/2203.05736.pdf.
How about using memories? Such as transformer recurrence