2

I want a fast offline speech recognizer (like vosk or sphinx) as a wake-up word for google speech recognition so it doesn't always hear what we say.

Want wake up word for this:

def takeCommand():
# mic input

r = sr.Recognizer()
with sr.Microphone() as source:
    print("Hearing...")
    # r.pause_threshold = 1
    r.chunk_size = 2048
    r.adjust_for_ambient_noise(source)
    audio = r.listen(source)

try:
    print("Recognizing....")
    query = r.recognize_google(audio, language='en-in')
    print(f"You said: {query}\n")

except Exception as e:
    print(e)

    print("Say that again please...")
    return "None"
return query
Snipey
  • 61
  • 2

1 Answers1

0

Vosk should work well for a fast wake-up word. Choose a wake-up word/phrase that is covered by its language model and doesn't come up frequently in normal conversation. Then use the partial callback for it to return a low-latency match.

You can get a quick idea of how well it will work by just browsing to https://ccoreilly.github.io/vosk-browser/ and trying out some example wake-up words with the web-based demo.

If you have a separate "awake" and "sleep" mode in your state management, you can probably be very liberal about adding homonyms and close-enough matching for your wake word. E.g. If your wake word is "wake up", you can also accept "way up" and other phrases as the wake word.

You should get sub-second matches on most platforms though hardware differences can change this significantly.

Erik Hermansen
  • 2,200
  • 3
  • 21
  • 41