A minimalist, performance-oriented inference server for automatic speech recognition. Here's some prelimary performance numbers. These numbers are for total round-trip request time, including ...