From c5e08b24090d34d91af6e0d6b2f340e23c33b3ae Mon Sep 17 00:00:00 2001
From: Denis Shulyaka <Shulyaka@gmail.com>
Date: Tue, 5 May 2026 17:28:20 +0300
Subject: [PATCH] Return the requested format for OpenAI TTS (#169839)

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
---
 .../components/openai_conversation/tts.py     | 21 +++---
 .../openai_conversation/test_tts.py           | 73 +++++++++++++++++++
 2 files changed, 84 insertions(+), 10 deletions(-)

diff --git a/homeassistant/components/openai_conversation/tts.py b/homeassistant/components/openai_conversation/tts.py
index f3ee614d7470..9aab49a465a5 100644
--- a/homeassistant/components/openai_conversation/tts.py
+++ b/homeassistant/components/openai_conversation/tts.py
@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from collections.abc import Mapping
 import logging
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING, Any, Literal
 
 from openai import OpenAIError
 from propcache.api import cached_property
@@ -166,14 +166,15 @@ class OpenAITTSEntity(TextToSpeechEntity, OpenAIBaseLLMEntity):
         client = self.entry.runtime_data
 
         response_format = options[ATTR_PREFERRED_FORMAT]
-        if response_format not in self._supported_formats:
-            # common aliases
-            if response_format == "ogg":
-                response_format = "opus"
-            elif response_format == "raw":
-                response_format = "pcm"
-            else:
-                response_format = self.default_options[ATTR_PREFERRED_FORMAT]
+        if response_format in ("ogg", "oga"):
+            codec: Literal["mp3", "opus", "aac", "flac", "wav", "pcm"] = "opus"
+        elif response_format == "raw":
+            response_format = codec = "pcm"
+        elif response_format not in self._supported_formats:
+            response_format = self.default_options[ATTR_PREFERRED_FORMAT]
+            codec = response_format
+        else:
+            codec = response_format
 
         try:
             async with client.audio.speech.with_streaming_response.create(
@@ -182,7 +183,7 @@ class OpenAITTSEntity(TextToSpeechEntity, OpenAIBaseLLMEntity):
                 input=message,
                 instructions=str(options.get(CONF_PROMPT)),
                 speed=options.get(CONF_TTS_SPEED, RECOMMENDED_TTS_SPEED),
-                response_format=response_format,
+                response_format=codec,
             ) as response:
                 response_data = bytearray()
                 async for chunk in response.iter_bytes():
diff --git a/tests/components/openai_conversation/test_tts.py b/tests/components/openai_conversation/test_tts.py
index ffbe7117c1e8..45c3a0a96204 100644
--- a/tests/components/openai_conversation/test_tts.py
+++ b/tests/components/openai_conversation/test_tts.py
@@ -118,6 +118,79 @@ async def test_tts(
     )
 
 
+@pytest.mark.parametrize(
+    ("preferred_format", "expected_response_format"),
+    [
+        ("ogg", "opus"),
+        ("oga", "opus"),
+        ("mp3", "mp3"),
+    ],
+)
+@pytest.mark.usefixtures("mock_init_component")
+async def test_tts_preferred_format(
+    hass: HomeAssistant,
+    hass_client: ClientSessionGenerator,
+    mock_create_speech: MagicMock,
+    calls: list[ServiceCall],
+    preferred_format: str,
+    expected_response_format: str,
+) -> None:
+    """Test text to speech preferred format handling."""
+    mock_create_speech.return_value = [b"mock audio data"]
+
+    await hass.services.async_call(
+        tts.DOMAIN,
+        "speak",
+        {
+            ATTR_ENTITY_ID: "tts.openai_tts",
+            tts.ATTR_MEDIA_PLAYER_ENTITY_ID: "media_player.something",
+            tts.ATTR_MESSAGE: "There is a person at the front door.",
+            tts.ATTR_OPTIONS: {tts.ATTR_PREFERRED_FORMAT: preferred_format},
+        },
+        blocking=True,
+    )
+
+    assert len(calls) == 1
+    assert (
+        await retrieve_media(hass, hass_client, calls[0].data[ATTR_MEDIA_CONTENT_ID])
+        == HTTPStatus.OK
+    )
+    mock_create_speech.assert_called_once_with(
+        model="gpt-4o-mini-tts",
+        voice="marin",
+        input="There is a person at the front door.",
+        instructions="",
+        speed=1.0,
+        response_format=expected_response_format,
+    )
+
+
+@pytest.mark.usefixtures("mock_init_component")
+async def test_tts_raw_preferred_format_returns_pcm(
+    hass: HomeAssistant,
+    mock_create_speech: MagicMock,
+) -> None:
+    """Test raw preferred format is returned as pcm."""
+    tts_entity = hass.data[tts.DOMAIN].get_entity("tts.openai_tts")
+    mock_create_speech.return_value = [b"mock audio data"]
+
+    result = await tts_entity.async_get_tts_audio(
+        "There is a person at the front door.",
+        "en-US",
+        {tts.ATTR_PREFERRED_FORMAT: "raw", tts.ATTR_VOICE: "marin"},
+    )
+
+    assert result == ("pcm", b"mock audio data")
+    mock_create_speech.assert_called_once_with(
+        model="gpt-4o-mini-tts",
+        voice="marin",
+        input="There is a person at the front door.",
+        instructions="",
+        speed=1.0,
+        response_format="pcm",
+    )
+
+
 @pytest.mark.usefixtures("mock_init_component")
 async def test_tts_error(
     hass: HomeAssistant,