Automatic speech recognition has recently received enormous attention with the advent of cloud-based services such as Cortana, Siri, Majel and other virtual avatars. A recording comes from the device, is processed on a remote server, which sends some text back to the user. Round-trip latency: <200ms. Can we ever achieve better results with a microprocessor running FOSS or is speech truly a black box problem? State of the art ASR libraries are certainly slower and less accurate, suggesting all of the innovation will happen inside a datacenter or microchip, behind NDAs and closed doors. We will review the state of offline ASR technology and attempt to deconstruct that notion. No prior speech recognition experience is required.