Yeah a small (ideally personalized) wakeword model would probably outperform just about any audio wizardry.