diff --git a/aider/voice.py b/aider/voice.py
index a9d5120dd..accff611e 100644
--- a/aider/voice.py
+++ b/aider/voice.py
@@ -141,18 +141,20 @@ class Voice:
             while not self.q.empty():
                 file.write(self.q.get())
 
+        use_audio_format = self.audio_format
+
         # Check file size and offer to convert to mp3 if too large
         file_size = os.path.getsize(temp_wav)
         if file_size > 24.9 * 1024 * 1024 and self.audio_format == "wav":
             print("\nWarning: {temp_wav} is too large, switching to mp3 format.")
-            self.audio_format = "mp3"
+            use_audio_format = "mp3"
 
         filename = temp_wav
-        if self.audio_format != "wav":
+        if use_audio_format != "wav":
             try:
-                new_filename = tempfile.mktemp(suffix=f".{self.audio_format}")
+                new_filename = tempfile.mktemp(suffix=f".{use_audio_format}")
                 audio = AudioSegment.from_wav(temp_wav)
-                audio.export(filename, format=self.audio_format)
+                audio.export(filename, format=use_audio_format)
                 os.remove(temp_wav)
                 filename = new_filename
             except (CouldntDecodeError, CouldntEncodeError) as e:
@@ -171,7 +173,7 @@ class Voice:
                 print(f"Unable to transcribe {filename}: {err}")
                 return
 
-        if self.audio_format != "wav":
+        if filename != temp_wav:
             os.remove(filename)
 
         text = transcript.text