From 079d8e32c99a899f80d88e38ea63678c56258edf Mon Sep 17 00:00:00 2001
From: Giacomo Saccaggi <44114264+GiacomoSaccaggi@users.noreply.github.com>
Date: Wed, 2 Sep 2026 11:16:51 +0200
Subject: [PATCH] Use iter_chunked for STT audio stream to prevent LineTooLong
 (#180831)

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
---
 homeassistant/components/stt/__init__.py |  9 +++++--
 tests/components/stt/test_init.py        | 30 ++++++++++++++++++++++++
 2 files changed, 37 insertions(+), 2 deletions(-)

diff --git a/homeassistant/components/stt/__init__.py b/homeassistant/components/stt/__init__.py
index c4ad6297868c..8a98763b31c4 100644
--- a/homeassistant/components/stt/__init__.py
+++ b/homeassistant/components/stt/__init__.py
@@ -72,6 +72,11 @@ _LOGGER = logging.getLogger(__name__)
 
 CONFIG_SCHEMA = cv.empty_config_schema(DOMAIN)
 
+# Audio is read from the request body in chunks of at most 4096 bytes because
+# line-based iteration raises LineTooLong on binary audio without newline bytes.
+# At 16 kHz/16-bit/mono, 4096 bytes represents up to 128 ms of audio.
+AUDIO_CHUNK_SIZE = 4096
+
 
 @callback
 def async_default_engine(hass: HomeAssistant) -> str | None:
@@ -291,7 +296,7 @@ class SpeechToTextView(HomeAssistantView):
 
             # Process audio stream
             result = await stt_provider.async_process_audio_stream(
-                metadata, request.content
+                metadata, request.content.iter_chunked(AUDIO_CHUNK_SIZE)
             )
         else:
             # Check format
@@ -300,7 +305,7 @@ class SpeechToTextView(HomeAssistantView):
 
             # Process audio stream
             result = await provider_entity.internal_async_process_audio_stream(
-                metadata, request.content
+                metadata, request.content.iter_chunked(AUDIO_CHUNK_SIZE)
             )
 
         # Return result
diff --git a/tests/components/stt/test_init.py b/tests/components/stt/test_init.py
index 8ff45b595779..76dc65329693 100644
--- a/tests/components/stt/test_init.py
+++ b/tests/components/stt/test_init.py
@@ -644,3 +644,33 @@ async def test_audio_processing_custom(hass: HomeAssistant, tmp_path: Path) -> N
     assert engine.audio_processing.requires_external_vad is False
     assert engine.audio_processing.prefers_auto_gain_enabled is False
     assert engine.audio_processing.prefers_noise_reduction_enabled is False
+
+
+@pytest.mark.parametrize(
+    "setup", ["mock_setup", "mock_config_entry_setup"], indirect=True
+)
+async def test_stream_audio_large_no_newline_block(
+    hass: HomeAssistant,
+    hass_client: ClientSessionGenerator,
+    setup: MockSTTProvider | MockSTTProviderEntity,
+) -> None:
+    """Test a newline-free audio stream larger than aiohttp's line limit."""
+    # 600,000 bytes of silence - larger than the 512KB LineTooLong threshold.
+    # This is the reproduction case from issue #180708.
+    test_data = b"\x00" * 600_000
+    client = await hass_client()
+    response = await client.post(
+        f"/api/stt/{setup.url_path}",
+        headers={
+            "X-Speech-Content": (
+                "format=wav; codec=pcm; sample_rate=16000; bit_rate=16; channel=1;"
+                " language=en"
+            )
+        },
+        data=test_data,
+    )
+    assert response.status == HTTPStatus.OK
+    assert await response.json() == {"text": "test_result", "result": "success"}
+
+    received_data = b"".join(setup.received)
+    assert received_data == test_data