Add TTS to pipelines (#90004)

* Add text to speech and stages to pipeline

* Default to "cloud" TTS when engine is None

* Refactor pipeline request to split text/audio

* Refactor with PipelineRun

* Generate pipeline from language

* Clean up

* Restore TTS code

* Add audio pipeline test

* Clean TTS cache in test

* Clean up tests and pipeline base class

* Stop pylint and pytest magics from fighting

* Include mock_get_cache_files

homeassistant/components/voice_assistant/__init__.py

@@ -4,20 +4,13 @@ from __future__ import annotations
 from homeassistant.core import HomeAssistant
 from homeassistant.helpers.typing import ConfigType
 
-from .const import DEFAULT_PIPELINE, DOMAIN
-from .pipeline import Pipeline
+from .const import DOMAIN
 from .websocket_api import async_register_websocket_api
 
 
 async def async_setup(hass: HomeAssistant, config: ConfigType) -> bool:
     """Set up Voice Assistant integration."""
-    hass.data[DOMAIN] = {
-        DEFAULT_PIPELINE: Pipeline(
-            name=DEFAULT_PIPELINE,
-            language=None,
-            conversation_engine=None,
-        )
-    }
+    hass.data[DOMAIN] = {}
     async_register_websocket_api(hass)
 
     return True

homeassistant/components/voice_assistant/pipeline.py

@@ -1,6 +1,7 @@
 """Classes for voice assistant pipelines."""
 from __future__ import annotations
 
+from abc import ABC, abstractmethod
 import asyncio
 from collections.abc import Callable
 from dataclasses import dataclass, field
@@ -8,20 +9,16 @@ from typing import Any
 
 from homeassistant.backports.enum import StrEnum
 from homeassistant.components import conversation
+from homeassistant.components.media_source import async_resolve_media
+from homeassistant.components.tts.media_source import (
+    generate_media_source_id as tts_generate_media_source_id,
+)
 from homeassistant.core import Context, HomeAssistant
 from homeassistant.util.dt import utcnow
 
 DEFAULT_TIMEOUT = 30  # seconds
 
 
-@dataclass
-class PipelineRequest:
-    """Request to start a pipeline run."""
-
-    intent_input: str
-    conversation_id: str | None = None
-
-
 class PipelineEventType(StrEnum):
     """Event types emitted during a pipeline run."""
 
@@ -29,6 +26,8 @@ class PipelineEventType(StrEnum):
     RUN_FINISH = "run-finish"
     INTENT_START = "intent-start"
     INTENT_FINISH = "intent-finish"
+    TTS_START = "tts-start"
+    TTS_FINISH = "tts-finish"
     ERROR = "error"
 
 
@@ -56,69 +55,161 @@ class Pipeline:
     name: str
     language: str | None
     conversation_engine: str | None
+    tts_engine: str | None
 
-    async def run(
-        self,
-        hass: HomeAssistant,
-        context: Context,
-        request: PipelineRequest,
-        event_callback: Callable[[PipelineEvent], None],
-        timeout: int | float | None = DEFAULT_TIMEOUT,
-    ) -> None:
-        """Run a pipeline with an optional timeout."""
-        await asyncio.wait_for(
-            self._run(hass, context, request, event_callback), timeout=timeout
-        )
 
-    async def _run(
-        self,
-        hass: HomeAssistant,
-        context: Context,
-        request: PipelineRequest,
-        event_callback: Callable[[PipelineEvent], None],
-    ) -> None:
-        """Run a pipeline."""
-        language = self.language or hass.config.language
-        event_callback(
+@dataclass
+class PipelineRun:
+    """Running context for a pipeline."""
+
+    hass: HomeAssistant
+    context: Context
+    pipeline: Pipeline
+    event_callback: Callable[[PipelineEvent], None]
+    language: str = None  # type: ignore[assignment]
+
+    def __post_init__(self):
+        """Set language for pipeline."""
+        self.language = self.pipeline.language or self.hass.config.language
+
+    def start(self):
+        """Emit run start event."""
+        self.event_callback(
             PipelineEvent(
                 PipelineEventType.RUN_START,
                 {
-                    "pipeline": self.name,
-                    "language": language,
+                    "pipeline": self.pipeline.name,
+                    "language": self.language,
                 },
             )
         )
 
-        intent_input = request.intent_input
+    def finish(self):
+        """Emit run finish event."""
+        self.event_callback(
+            PipelineEvent(
+                PipelineEventType.RUN_FINISH,
+            )
+        )
 
-        event_callback(
+    async def recognize_intent(
+        self, intent_input: str, conversation_id: str | None
+    ) -> conversation.ConversationResult:
+        """Run intent recognition portion of pipeline."""
+        self.event_callback(
             PipelineEvent(
                 PipelineEventType.INTENT_START,
                 {
-                    "engine": self.conversation_engine or "default",
+                    "engine": self.pipeline.conversation_engine or "default",
                     "intent_input": intent_input,
                 },
             )
         )
 
         conversation_result = await conversation.async_converse(
-            hass=hass,
+            hass=self.hass,
             text=intent_input,
-            conversation_id=request.conversation_id,
-            context=context,
-            language=language,
-            agent_id=self.conversation_engine,
+            conversation_id=conversation_id,
+            context=self.context,
+            language=self.language,
+            agent_id=self.pipeline.conversation_engine,
         )
 
-        event_callback(
+        self.event_callback(
             PipelineEvent(
                 PipelineEventType.INTENT_FINISH,
                 {"intent_output": conversation_result.as_dict()},
             )
         )
 
-        event_callback(
+        return conversation_result
+
+    async def text_to_speech(self, tts_input: str) -> str:
+        """Run text to speech portion of pipeline. Returns URL of TTS audio."""
+        self.event_callback(
             PipelineEvent(
-                PipelineEventType.RUN_FINISH,
+                PipelineEventType.TTS_START,
+                {
+                    "engine": self.pipeline.tts_engine or "default",
+                    "tts_input": tts_input,
+                },
             )
         )
+
+        tts_media = await async_resolve_media(
+            self.hass,
+            tts_generate_media_source_id(
+                self.hass,
+                tts_input,
+                engine=self.pipeline.tts_engine,
+            ),
+        )
+        tts_url = tts_media.url
+
+        self.event_callback(
+            PipelineEvent(
+                PipelineEventType.TTS_FINISH,
+                {"tts_output": tts_url},
+            )
+        )
+
+        return tts_url
+
+
+@dataclass
+class PipelineRequest(ABC):
+    """Request to for a pipeline run."""
+
+    async def execute(
+        self, run: PipelineRun, timeout: int | float | None = DEFAULT_TIMEOUT
+    ):
+        """Run pipeline with optional timeout."""
+        await asyncio.wait_for(
+            self._execute(run),
+            timeout=timeout,
+        )
+
+    @abstractmethod
+    async def _execute(self, run: PipelineRun):
+        """Run pipeline with request info and context."""
+
+
+@dataclass
+class TextPipelineRequest(PipelineRequest):
+    """Request to run the text portion only of a pipeline."""
+
+    intent_input: str
+    conversation_id: str | None = None
+
+    async def _execute(
+        self,
+        run: PipelineRun,
+    ):
+        run.start()
+        await run.recognize_intent(self.intent_input, self.conversation_id)
+        run.finish()
+
+
+@dataclass
+class AudioPipelineRequest(PipelineRequest):
+    """Request to full pipeline from audio input (stt) to audio output (tts)."""
+
+    intent_input: str  # this will be changed to stt audio
+    conversation_id: str | None = None
+
+    async def _execute(self, run: PipelineRun):
+        run.start()
+
+        # stt will go here
+
+        conversation_result = await run.recognize_intent(
+            self.intent_input, self.conversation_id
+        )
+
+        tts_input = conversation_result.response.speech.get("plain", {}).get(
+            "speech", ""
+        )
+
+        await run.text_to_speech(tts_input)
+
+        run.finish()

homeassistant/components/voice_assistant/websocket_api.py

@@ -7,7 +7,7 @@ from homeassistant.components import websocket_api
 from homeassistant.core import HomeAssistant, callback
 
 from .const import DOMAIN
-from .pipeline import DEFAULT_TIMEOUT, PipelineRequest
+from .pipeline import DEFAULT_TIMEOUT, Pipeline, PipelineRun, TextPipelineRequest
 
 
 @callback
@@ -19,7 +19,8 @@ def async_register_websocket_api(hass: HomeAssistant) -> None:
 @websocket_api.websocket_command(
     {
         vol.Required("type"): "voice_assistant/run",
-        vol.Optional("pipeline", default="default"): str,
+        vol.Optional("language"): str,
+        vol.Optional("pipeline"): str,
         vol.Required("intent_input"): str,
         vol.Optional("conversation_id"): vol.Any(str, None),
         vol.Optional("timeout"): vol.Any(float, int),
@@ -32,28 +33,43 @@ async def websocket_run(
     msg: dict[str, Any],
 ) -> None:
     """Run a pipeline."""
-    pipeline_id = msg["pipeline"]
+    pipeline_id = msg.get("pipeline")
+    if pipeline_id is not None:
         pipeline = hass.data[DOMAIN].get(pipeline_id)
         if pipeline is None:
             connection.send_error(
-            msg["id"], "pipeline_not_found", f"Pipeline not found: {pipeline_id}"
+                msg["id"],
+                "pipeline_not_found",
+                f"Pipeline not found: {pipeline_id}",
             )
             return
 
+    else:
+        # Construct a pipeline for the required/configured language
+        language = msg.get("language", hass.config.language)
+        pipeline = Pipeline(
+            name=language,
+            language=language,
+            conversation_engine=None,
+            tts_engine=None,
+        )
+
     # Run pipeline with a timeout.
     # Events are sent over the websocket connection.
     timeout = msg.get("timeout", DEFAULT_TIMEOUT)
     run_task = hass.async_create_task(
-        pipeline.run(
-            hass,
-            connection.context(msg),
-            request=PipelineRequest(
+        TextPipelineRequest(
             intent_input=msg["intent_input"],
             conversation_id=msg.get("conversation_id"),
-            ),
+        ).execute(
+            PipelineRun(
+                hass,
+                connection.context(msg),
+                pipeline,
                 event_callback=lambda event: connection.send_event(
                     msg["id"], event.as_dict()
                 ),
+            ),
             timeout=timeout,
         )
     )

tests/components/voice_assistant/test_pipeline.py

@@ -0,0 +1,110 @@
+"""Pipeline tests for Voice Assistant integration."""
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from homeassistant.components.voice_assistant.pipeline import (
+    AudioPipelineRequest,
+    Pipeline,
+    PipelineEventType,
+    PipelineRun,
+)
+from homeassistant.core import Context
+from homeassistant.setup import async_setup_component
+
+from tests.components.tts.conftest import (  # noqa: F401, pylint: disable=unused-import
+    mock_get_cache_files,
+    mock_init_cache_dir,
+)
+
+
+@pytest.fixture(autouse=True)
+async def init_components(hass):
+    """Initialize relevant components with empty configs."""
+    assert await async_setup_component(hass, "voice_assistant", {})
+
+
+@pytest.fixture
+async def mock_get_tts_audio(hass):
+    """Set up media source."""
+    assert await async_setup_component(hass, "media_source", {})
+    assert await async_setup_component(
+        hass,
+        "tts",
+        {
+            "tts": {
+                "platform": "demo",
+            }
+        },
+    )
+
+    with patch(
+        "homeassistant.components.demo.tts.DemoProvider.get_tts_audio",
+        return_value=("mp3", b""),
+    ) as mock_get_tts:
+        yield mock_get_tts
+
+
+async def test_audio_pipeline(hass, mock_get_tts_audio):
+    """Run audio pipeline with mock TTS."""
+    pipeline = Pipeline(
+        name="test",
+        language=hass.config.language,
+        conversation_engine=None,
+        tts_engine=None,
+    )
+
+    event_callback = MagicMock()
+    await AudioPipelineRequest(intent_input="Are the lights on?").execute(
+        PipelineRun(
+            hass,
+            context=Context(),
+            pipeline=pipeline,
+            event_callback=event_callback,
+            language=hass.config.language,
+        )
+    )
+
+    calls = event_callback.mock_calls
+    assert calls[0].args[0].type == PipelineEventType.RUN_START
+    assert calls[0].args[0].data == {
+        "pipeline": "test",
+        "language": hass.config.language,
+    }
+
+    assert calls[1].args[0].type == PipelineEventType.INTENT_START
+    assert calls[1].args[0].data == {
+        "engine": "default",
+        "intent_input": "Are the lights on?",
+    }
+    assert calls[2].args[0].type == PipelineEventType.INTENT_FINISH
+    assert calls[2].args[0].data == {
+        "intent_output": {
+            "conversation_id": None,
+            "response": {
+                "card": {},
+                "data": {"code": "no_intent_match"},
+                "language": hass.config.language,
+                "response_type": "error",
+                "speech": {
+                    "plain": {
+                        "extra_data": None,
+                        "speech": "Sorry, I couldn't understand that",
+                    }
+                },
+            },
+        }
+    }
+
+    assert calls[3].args[0].type == PipelineEventType.TTS_START
+    assert calls[3].args[0].data == {
+        "engine": "default",
+        "tts_input": "Sorry, I couldn't understand that",
+    }
+    assert calls[4].args[0].type == PipelineEventType.TTS_FINISH
+    assert (
+        calls[4].args[0].data["tts_output"]
+        == f"/api/tts_proxy/dae2cdcb27a1d1c3b07ba2c7db91480f9d4bfd8f_{hass.config.language}_-_demo.mp3"
+    )
+
+    assert calls[5].args[0].type == PipelineEventType.RUN_FINISH

tests/components/voice_assistant/test_websocket.py

@@ -24,7 +24,11 @@ async def test_text_only_pipeline(
     client = await hass_ws_client(hass)
 
     await client.send_json(
-        {"id": 5, "type": "voice_assistant/run", "intent_input": "Are the lights on?"}
+        {
+            "id": 5,
+            "type": "voice_assistant/run",
+            "intent_input": "Are the lights on?",
+        }
     )
 
     # result
@@ -35,7 +39,7 @@ async def test_text_only_pipeline(
     msg = await client.receive_json()
     assert msg["event"]["type"] == "run-start"
     assert msg["event"]["data"] == {
-        "pipeline": "default",
+        "pipeline": hass.config.language,
         "language": hass.config.language,
     }
 
@@ -83,7 +87,8 @@ async def test_conversation_timeout(
         await asyncio.sleep(3600)
 
     with patch(
-        "homeassistant.components.conversation.async_converse", new=sleepy_converse
+        "homeassistant.components.conversation.async_converse",
+        new=sleepy_converse,
     ):
         await client.send_json(
             {
@@ -102,7 +107,7 @@ async def test_conversation_timeout(
         msg = await client.receive_json()
         assert msg["event"]["type"] == "run-start"
         assert msg["event"]["data"] == {
-            "pipeline": "default",
+            "pipeline": hass.config.language,
             "language": hass.config.language,
         }
 
@@ -130,7 +135,7 @@ async def test_pipeline_timeout(
         await asyncio.sleep(3600)
 
     with patch(
-        "homeassistant.components.voice_assistant.pipeline.Pipeline._run",
+        "homeassistant.components.voice_assistant.pipeline.TextPipelineRequest._execute",
         new=sleepy_run,
     ):
         await client.send_json(