"""Sub-million-parameter streaming speech recognition experiments."""