Raspberry Pi Asynchronous/Continuous Speech Recognition in Python

Question

I want to create a speech recognition script for the Raspberry Pi in Python and need an asynchronous/continuous speech recognition library. Asynchronous means that I need endless running of the recognition until the spoken matches to an array of words without any input from a keyboard, and then display the spoken to the terminal and restart recognition. I already had a look at PocketSphinx, but after a few hours Googling, I didn't find anything about an Asynchronous recognition with that.

Do you know any Library who is capable of that?

Answer 1

You can use Pocketsphinx on Raspberry Pi. You need to download latest version 5prealpha.

It can listen for multiple keyphrases. The code should be something like this:

import sys, os
from pocketsphinx import *
import pyaudio

modeldir = "../../../model"

# Create a decoder with certain model
config = Decoder.default_config()
config.set_string('-hmm', os.path.join(modeldir, 'en-us/en-us'))
config.set_string('-dict', os.path.join(modeldir, 'en-us/cmudict-en-us.dict'))
config.set_string('-kws', 'keyphrase.list')

p = pyaudio.PyAudio()
stream = p.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True, frames_per_buffer=1024)
stream.start_stream()

# Process audio chunk by chunk. On keyword detected perform action and restart search
decoder = Decoder(config)
decoder.start_utt()
while True:
    buf = stream.read(1024)
    decoder.process_raw(buf, False, False)
    if decoder.hyp() != None:
        print "Detected keyword", decoder.hyp(), "restarting search"
        decoder.end_utt()
        decoder.start_utt()

The keypharse.list file should look like this, one phrase per line with threshold

open the door /1e-40/
close the door /1e-40/
how are you /1e-30/

Thresholds must be tuned for every keyphrase to balance between false alarms and misdetections.