diff --git a/src/whisper_ctranslate2/transcribe.py b/src/whisper_ctranslate2/transcribe.py
index 2225166..b8a4c24 100644
--- a/src/whisper_ctranslate2/transcribe.py
+++ b/src/whisper_ctranslate2/transcribe.py
@@ -122,13 +122,10 @@ def __init__(
             download_root=cache_directory,
             local_files_only=local_files_only,
         )
+
+        self.batch_size = batch_size
         if batched:
-            if batch_size:
-                self.batched_model = BatchedInferencePipeline(model=self.model)
-            else:
-                self.batched_model = BatchedInferencePipeline(
-                    model=self.model, batch_size=batch_size
-                )
+            self.batched_model = BatchedInferencePipeline(model=self.model)
         else:
             self.batched_model = None
 
@@ -150,6 +147,10 @@ def inference(
             model = self.model
             vad = options.vad_filter
 
+        batch_size = (
+            {"batch_size": self.batch_size} if self.batch_size is not None else {}
+        )
+
         segments, info = model.transcribe(
             audio=audio,
             language=language,
@@ -177,6 +178,7 @@ def inference(
             hallucination_silence_threshold=options.hallucination_silence_threshold,
             vad_filter=vad,
             vad_parameters=vad_parameters,
+            **batch_size,
         )
 
         language_name = LANGUAGES[info.language].title()