DSP Group’s DBMD4D is an innovative voice processor enabling voice user interface in IoT devices and supporting device activation and operation via voice commands while maintaining extremely low ...

Cadence Tensilica HiFi 5 DSP for audio/voice is the first IP core for high-performance far-field processing/AI-based speech recognition processing.

That datasheet’s SoC diagram shows an audio DSP, too. We haven’t heard anything about speech recognition related to the Atom’s Smart Sound implementation.