Memory leak problem still existed in version 2.0.10

[MLrookie]

Memory leak problem still existed in version 2.0.10, but it is said thath it has been fixed in 2.0.10 according to README.md. If you run the following code, you will found that the memory of process quickly increase.

for _ in range(100000):
        webrtcvad.Vad(3)

[yaleimeng]

one Vad() object can deal many voice files， you don't need to create so many Vad() objetcs.
According to the cpu cores and threads, I think 8~32 is enough.

[willnufe]

This seems to be due to the issue of the vad_free function not being called properly in the pywebrtcvad.c file.
I made modifications to the code for testing.

  static void vad_free(PyObject* vadptr)
  {
    VadInst* handle = PyCapsule_GetPointer(vadptr, "WebRtcVadPtr");
    printf("vad_free\n");
    WebRtcVad_Free(handle);
  }
  
  static PyObject* vad_create(PyObject *self, PyObject *args)
  {
    VadInst *handle = WebRtcVad_Create();
    printf("vad_create\n");
    // PyObject *vadptr = PyCapsule_New(handle, "WebRtcVadPtr", vad_free);
    // return Py_BuildValue("O", vadptr);
    return PyCapsule_New(handle, "WebRtcVadPtr", vad_free);
  }

# testing
import webrtcvad
from tqdm import tqdm

for _ in tqdm(range(10)):
    print("im here begin")
    vad = webrtcvad.Vad(3)
    print("im here end")

    del vad  # This line of code needs to be added to free up memory
    print("im here del end")
    print("\n\n")

# res
"""
im here begin
vad_create
im here end
vad_free
im here del end

...
"""

If it is the original code, it will not print the 'vad_free', which means it will not release memory normally.
But I am not sure if this is the cause of the memory leak. Can someone provide verification?

[arkadijs]

Hi, any updates on this and PR #49?
Pooling VAD instances sometimes is not an option as reusing them across streams produces non-repeatable results, unfortunately.

[daanzu]

FYI, I have implemented this change in my fork at https://github.com/daanzu/py-webrtcvad-wheels .