Hacker News Reader

Does anyone know of papers and/or example implementations of similar DNNs for accoustic modeling using tf or some other framework?

Google's Deep KWS paper [1] is kind of similar, although they don't use an HMM.