How do voice assistants process commands?

12 Tem 00:42

Voice assistants essentially divide the processing between local chips and the cloud. Take Siri or Google Assistant, for example—they perform initial processing on-device using neural engines (like Apple’s A13/A14/A15/A16 Bionic or Qualcomm’s Hexagon in Snapdragons) to quickly detect keywords and suppress noise. That first 100–150 ms of latency you notice is usually the local step; it filters out the “Hey Siri” so the rest of the query can be sent to the cloud without wasting bandwidth. I’ve seen offline commands reach ~92% accuracy for native speakers on my Pixel 6, but once you move into broad Glaswegian or thick Boston accents, you’re looking at mid-80% accuracy unless you’ve trained a profile. Accuracy for non-native speakers or heavy accents depends on how much the cloud model has been exposed to. If you say something like “set an alarm for 5:30 AM,” it’ll still pick up the numbers and time intent, but plural nouns (“five thirty”) can throw it off. For accents, I’ve found a simple trick: before speaking, switch the assistant’s language to match your phonetic roots—for example, set Assistant to English (UK) if you’ve got a Scouse burr; the cloud model then adjusts the linguistic pipeline toward British English acoustics. You won’t hit 99.9%, but you’ll recover 3–5% accuracy points without extra cloud cycles.