feat(types): add typed support for Speech Understanding action_items request and response

assemblyai/__init__.py

@@ -52,6 +52,9 @@
     TTSWordBoundariesEvent,
 )
 from .types import (
+    ActionItem,
+    ActionItemsRequest,
+    ActionItemsResponse,
     AssemblyAIError,
     AutohighlightResponse,
     AutohighlightResult,
@@ -145,6 +148,9 @@ def __getattr__(name: str):
 
 __all__ = [
     # types
+    "ActionItem",
+    "ActionItemsRequest",
+    "ActionItemsResponse",
     "AssemblyAIError",
     "AsyncClient",
     "AsyncDictationTranscriber",

assemblyai/__version__.py

@@ -1 +1 @@
-__version__ = "1.9.0"
+__version__ = "1.9.1"

assemblyai/types.py

@@ -773,6 +773,18 @@ class SummarizationRequest(BaseModel):
     "How much effort to spend on this task (default: 'low')"
 
 
+class ActionItemsRequest(BaseModel):
+    """
+    Action items configuration for speech understanding
+    """
+
+    include_decisions: Optional[bool] = None
+    "Also extract key decisions made during the conversation"
+
+    effort: Optional[Literal["low", "medium"]] = None
+    "How much effort to spend on this task (default: 'low')"
+
+
 class SpeechUnderstandingFeatureRequests(BaseModel):
     """
     Speech understanding feature requests
@@ -790,6 +802,9 @@ class SpeechUnderstandingFeatureRequests(BaseModel):
     summarization: Optional[SummarizationRequest] = None
     "Summarization configuration"
 
+    action_items: Optional[ActionItemsRequest] = None
+    "Action items configuration"
+
 
 class SpeechUnderstandingRequest(BaseModel):
     """
@@ -890,6 +905,36 @@ class SummarizationResponse(BaseModel):
     "The effort level that was used"
 
 
+class ActionItem(BaseModel):
+    """
+    An action item or decision extracted from the audio
+    """
+
+    action_item: str
+    "The extracted action item or decision"
+
+    quote: Optional[str] = None
+    "The source quote from the audio"
+
+    timestamp: Optional[int] = None
+    "Timestamp in milliseconds from audio start"
+
+
+class ActionItemsResponse(BaseModel):
+    """
+    Action items response containing status and extracted items
+    """
+
+    status: str
+    "Status of the action items feature"
+
+    effort: Optional[Literal["low", "medium"]] = None
+    "The effort level that was used"
+
+    items: Optional[List[ActionItem]] = None
+    "List of extracted action items"
+
+
 class SpeechUnderstandingFeatureResponses(BaseModel):
     """
     Speech understanding feature responses grouped together
@@ -907,6 +952,9 @@ class SpeechUnderstandingFeatureResponses(BaseModel):
     summarization: Optional[SummarizationResponse] = None
     "Summarization results"
 
+    action_items: Optional[ActionItemsResponse] = None
+    "Action items results including status and items"
+
 
 class SpeechUnderstandingResponse(BaseModel):
     """
@@ -1253,7 +1301,7 @@ def __init__(
             language_detection_options: Options for controlling the behavior or Automatic Language Detection.
             speech_threshold: Reject audio files that contain less than this fraction of speech. Valid values are in the range [0,1] inclusive.
             raw_transcription_config: Create the config from a `RawTranscriptionConfig`
-            speech_understanding: Speech understanding configuration for LLM Gateway features (speaker identification, translation, custom formatting)
+            speech_understanding: Speech understanding configuration for LLM Gateway features (speaker identification, translation, custom formatting, summarization, action items)
             temperature: Change how deterministic the response is, with 0 being the most deterministic and 1 being the least deterministic.
             remove_audio_tags: When set to 'all', removes all bracketed audio/speaker tags from the transcript. Only supported for Universal-3 Pro.
             keyterms_prompt_options: Options for controlling keyterms boosting behavior when using `keyterms_prompt`.

tests/unit/test_action_items.py

@@ -0,0 +1,138 @@
+from pytest_httpx import HTTPXMock
+
+import assemblyai as aai
+import tests.unit.factories as factories
+import tests.unit.unit_test_utils as test_utils
+
+aai.settings.api_key = "test"
+
+
+def test_action_items_request_model():
+    req = aai.ActionItemsRequest(
+        effort="low",
+        include_decisions=True,
+    )
+    assert req.effort == "low"
+    assert req.include_decisions is True
+
+
+def test_action_items_request_defaults():
+    req = aai.ActionItemsRequest()
+    assert req.effort is None
+    assert req.include_decisions is None
+
+
+def test_action_items_in_speech_understanding_request():
+    req = aai.SpeechUnderstandingRequest(
+        request=aai.SpeechUnderstandingFeatureRequests(
+            action_items=aai.ActionItemsRequest(
+                effort="medium",
+                include_decisions=False,
+            )
+        )
+    )
+    assert req.request is not None
+    assert req.request.action_items is not None
+    assert req.request.action_items.effort == "medium"
+    assert req.request.action_items.include_decisions is False
+
+
+def test_action_items_dict_style_in_transcription_config():
+    config = aai.TranscriptionConfig(
+        speech_understanding={
+            "request": {
+                "action_items": {
+                    "effort": "low",
+                    "include_decisions": True,
+                }
+            }
+        }
+    )
+    assert config.speech_understanding is not None
+    assert config.speech_understanding.request is not None
+    assert config.speech_understanding.request.action_items is not None
+    assert config.speech_understanding.request.action_items.effort == "low"
+    assert config.speech_understanding.request.action_items.include_decisions is True
+
+
+def test_action_items_response_model():
+    resp = aai.ActionItemsResponse(
+        status="success",
+        effort="low",
+        items=[
+            aai.ActionItem(
+                action_item="Send follow-up email",
+                quote="I will send follow-up email",
+                timestamp=124500,
+            )
+        ],
+    )
+    assert resp.status == "success"
+    assert resp.effort == "low"
+    assert resp.items is not None
+    assert len(resp.items) == 1
+    assert resp.items[0].action_item == "Send follow-up email"
+    assert resp.items[0].quote == "I will send follow-up email"
+    assert resp.items[0].timestamp == 124500
+
+
+def test_transcript_speech_understanding_action_items(httpx_mock: HTTPXMock):
+    mock_response = factories.generate_dict_factory(
+        factories.TranscriptCompletedResponseFactory
+    )()
+    mock_response["speech_understanding"] = {
+        "request": {
+            "action_items": {
+                "effort": "low",
+                "include_decisions": True,
+            }
+        },
+        "response": {
+            "action_items": {
+                "status": "success",
+                "effort": "low",
+                "items": [
+                    {
+                        "action_item": "Schedule sync meeting",
+                        "quote": "Let's meet tomorrow at 10",
+                        "timestamp": 45000,
+                    }
+                ],
+            }
+        },
+    }
+
+    request_body, transcript = test_utils.submit_mock_transcription_request(
+        httpx_mock,
+        mock_response,
+        config=aai.TranscriptionConfig(
+            speech_understanding={
+                "request": {
+                    "action_items": {
+                        "effort": "low",
+                        "include_decisions": True,
+                    }
+                }
+            }
+        ),
+    )
+
+    # Check request payload serialization
+    assert request_body.get("speech_understanding") is not None
+    su_request = request_body["speech_understanding"]["request"]
+    assert su_request.get("action_items") is not None
+    assert su_request["action_items"]["effort"] == "low"
+    assert su_request["action_items"]["include_decisions"] is True
+
+    # Check parsed transcript response
+    assert transcript.speech_understanding is not None
+    assert transcript.speech_understanding.response is not None
+    assert transcript.speech_understanding.response.action_items is not None
+    ai_resp = transcript.speech_understanding.response.action_items
+    assert ai_resp.status == "success"
+    assert ai_resp.effort == "low"
+    assert ai_resp.items is not None
+    assert len(ai_resp.items) == 1
+    assert ai_resp.items[0].action_item == "Schedule sync meeting"
+    assert ai_resp.items[0].quote == "Let's meet tomorrow at 10"
+    assert ai_resp.items[0].timestamp == 45000