Memory leak for very long stream

[nshmyrev]

As reported by Sebastian Dunnink

i've been trying out your vosk api and it's working great so far, very
easy to install and use. however i came across one problem, which is
that after about 20 seconds of runtime (a script based on your
microphone example) it uses up one more megabyte of ram. i found it by
chance while trying to understand why it stopped after several hours
of runtime. below is my script, i run it inside a screen instance.

[MikeyBeez]

You need to close the stream.

[dpny518]

I want to leave the stream open indefinitely, my goal is to use the asr as wake word system, and when it recognizes certain words trigger an action.
Any suggestions on how to use offline asr with continuous stream running indefinitely and avoid memory leaks?

[MikeyBeez]

Take a look at my program, Juliet. https://github.com/MikeyBeez/Juliet

[dpny518]

@MikeyBeez thank you, i checked it out, it seems you just check the output to see if a string is in it.

 while True:
        # listen for command. Speech to text listener logic is called from inside the myCommand function.
        output = mycommand.myCommand()[3:]
        # Remember,  the mycommand function takes in 
        # audio from the microphone and returns text.
        # Therefore, the "output" variable is text.
        if 'juli' in output:
            print('Julia responds:\n')

But I don't see any code for resetting, or handling memory issues

[MikeyBeez]

import pyaudio
from vosk import Model, KaldiRecognizer

def myCommand():
# "listens for commands"
# We imported vosk up above.
p = pyaudio.PyAudio()
stream = p.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True, frames_per_buffer=8000)
stream.start_stream()
model = Model("model-en")
rec = KaldiRecognizer(model, 16000)
while True:
data = stream.read(2000)
if len(data) == 0:
break
if rec.AcceptWaveform(data):
#print(rec.Result())
# I commented out this line and added the 3 lines below
myResult = rec.Result()
myList = myResult.split("text")
command = myList[1]
stream.stop_stream()
stream.close()
p.terminate()
return command

[MikeyBeez]

I rewrote this: from vosk import Model, KaldiRecognizer

import pyaudio

p = pyaudio.PyAudio()
stream = p.open(format=pyaudio.paInt16, channels=1, rate=16000, input=True, frames_per_buffer=8000)

model = Model("model-en")
rec = KaldiRecognizer(model, 16000)

while True:
stream.start_stream()
data = stream.read(2000)
if len(data) == 0:
break
if rec.AcceptWaveform(data):
#print(rec.Result())
# I commented out this line and added the 3 lines below
myResult = rec.Result()
myList = myResult.split("text")
print(myList[1])
stream.stop_stream()

[nshmyrev]

This should be fixed in commit ee9bacb and version 0.3.8