vad.py

import multiprocessing
from concurrent.futures import ProcessPoolExecutor, as_completed
# !pip install -q torchaudio
SAMPLING_RATE = 16000
import torch
from pprint import pprint

torch.set_num_threads(1)
NUM_PROCESS=4 # set to the number of CPU cores in the machine


model, utils = torch.hub.load(repo_or_dir='snakers4/silero-vad',
                                model='silero_vad',
                                force_reload=True,
                                onnx=False)

(get_speech_timestamps,
save_audio,
read_audio,
VADIterator,
collect_chunks) = utils

vad_models = dict()

def init_model(model):
  pid = multiprocessing.current_process().pid
  model, _ = torch.hub.load(repo_or_dir='snakers4/silero-vad',
                                model='silero_vad',
                                force_reload=False,
                                onnx=False)
  vad_models[pid] = model

def vad_process(audio_file: str):
    pid = multiprocessing.current_process().pid
    with torch.no_grad():
        wav = read_audio(audio_file, sampling_rate=SAMPLING_RATE)
        return get_speech_timestamps(
            wav,
            vad_models[pid],
            0.46,  # speech prob threshold
            16000,  # sample rate
            300,  # min speech duration in ms
            20,  # max speech duration in seconds
            600,  # min silence duration
            512,  # window size
            200,  # spech pad ms
        )
    

futures = []
with ProcessPoolExecutor(max_workers=NUM_PROCESS, initializer=init_model, initargs=(model,)) as ex:
    for i in ["file1.wav","file2.wav"]:
        futures.append(ex.submit(vad_process, i))

for finished in as_completed(futures):
    pprint(finished.result())