diff --git a/homeassistant/components/tts/__init__.py b/homeassistant/components/tts/__init__.py
index 1262e8132131..fc5fe4ffe23c 100644
--- a/homeassistant/components/tts/__init__.py
+++ b/homeassistant/components/tts/__init__.py
@@ -355,6 +355,9 @@ async def _async_convert_audio(
     if to_sample_bytes == 2:
         # 16-bit samples.
         command.extend(["-sample_fmt", "s16"])
+    # Do not write the muxer's own encoder metadata; metadata from the input
+    # is still copied.
+    command.extend(["-fflags", "+bitexact"])
     command.append("pipe:1")  # Send output to stdout.
 
     process = await asyncio.create_subprocess_exec(
diff --git a/tests/components/tts/test_init.py b/tests/components/tts/test_init.py
index c03cc2b380ab..3382c63f69b0 100644
--- a/tests/components/tts/test_init.py
+++ b/tests/components/tts/test_init.py
@@ -1920,6 +1920,8 @@ async def test_async_convert_audio_probe_size(
         "1",
         "-sample_fmt",
         "s16",
+        "-fflags",
+        "+bitexact",
         "pipe:1",
     ]
 
@@ -1968,6 +1970,8 @@ async def test_async_convert_audio_mp3_bitrate(
         "-ac",
         "1",
         *expected_encoder_args,
+        "-fflags",
+        "+bitexact",
         "pipe:1",
     ]