From 01340c20c7233ce226b94c537045f65d6baee4a6 Mon Sep 17 00:00:00 2001
From: Purfview <69023953+Purfview@users.noreply.github.com>
Date: Tue, 10 Dec 2024 18:35:02 +0000
Subject: [PATCH] Fixes OOM Errors - too high RAM usage by VAD

Reported problems:
https://github.com/SYSTRAN/faster-whisper/issues/1193
https://github.com/SYSTRAN/faster-whisper/issues/1169

VAD implementations consumes humongous memory amounts [original Silero doesn't have this problem]

This PR should fix the OOM problem.
Alt solution could be removing 'lru_cache'.
---
 faster_whisper/vad.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/faster_whisper/vad.py b/faster_whisper/vad.py
index 9605931c..fa790acf 100644
--- a/faster_whisper/vad.py
+++ b/faster_whisper/vad.py
@@ -260,8 +260,9 @@ def __init__(self, encoder_path, decoder_path):
             ) from e
 
         opts = onnxruntime.SessionOptions()
-        opts.inter_op_num_threads = 0
-        opts.intra_op_num_threads = 0
+        opts.inter_op_num_threads = 1
+        opts.intra_op_num_threads = 1
+        opts.enable_cpu_mem_arena = False
         opts.log_severity_level = 4
 
         self.encoder_session = onnxruntime.InferenceSession(