I guess you were talking about training a streaming ASR system with limited input look-ahead context? The default training config uses a transformer structure with full context, therefore it's not a model that can be used in a streaming system. Later on, we will release a few models that can work in streaming mode.
I guess you were talking about training a streaming ASR system with limited input look-ahead context? The default training config uses a transformer structure with full context, therefore it's not a model that can be used in a streaming system. Later on, we will release a few models that can work in streaming mode.