fix(types): keep raw_transcription_config values and rebuild transcript.config from them

assemblyai/prerecorded/v2/_base.py

@@ -100,10 +100,15 @@ def config_from_response(
 ) -> types.TranscriptionConfig:
     """Rebuilds the `TranscriptionConfig` a transcript was created with."""
 
+    # Goes through `RawTranscriptionConfig`: the response carries fields in their
+    # API shape (e.g. `custom_spelling` as a list of from/to objects) and fields
+    # that are not constructor arguments (e.g. `language_detection_results`).
     return types.TranscriptionConfig(
-        **response.dict(
-            include=set(types.RawTranscriptionConfig.__fields__),
-            exclude_none=True,
+        raw_transcription_config=types.RawTranscriptionConfig(
+            **response.dict(
+                include=set(types.RawTranscriptionConfig.__fields__),
+                exclude_none=True,
+            )
         )
     )
 

assemblyai/types.py

@@ -1259,13 +1259,11 @@ def __init__(
             keyterms_prompt_options: Options for controlling keyterms boosting behavior when using `keyterms_prompt`.
             domain: The domain to use for the transcription (e.g. 'medical-v1').
         """
-        self._raw_transcription_config = (
-            raw_transcription_config
-            if raw_transcription_config is not None
-            else RawTranscriptionConfig()
-        )
+        # The explicit arguments are applied to a fresh config first. Applying them
+        # straight onto `raw_transcription_config` would overwrite each of its
+        # values with the argument's `None` default.
+        self._raw_transcription_config = RawTranscriptionConfig()
 
-        # explicit configurations have higher priority if `raw_transcription_config` has been passed as well
         self.language_code = language_code
         self.language_codes = language_codes
         self.punctuate = punctuate
@@ -1320,6 +1318,13 @@ def __init__(
         self.speech_understanding = speech_understanding  # type: ignore[assignment]
         self.domain = domain
 
+        if raw_transcription_config is not None:
+            # explicit configurations have higher priority if `raw_transcription_config` has been passed as well
+            for field, value in self._raw_transcription_config:
+                if value is not None:
+                    setattr(raw_transcription_config, field, value)
+            self._raw_transcription_config = raw_transcription_config
+
     @property
     def raw(self) -> RawTranscriptionConfig:
         return self._raw_transcription_config

tests/unit/test_config.py

@@ -18,3 +18,37 @@ def test_configuration_are_none_by_default():
             pytest.fail(
                 f"Configuration field {name} is {value} and not None by default."
             )
+
+
+def test_raw_transcription_config_values_are_kept():
+    """
+    Tests that the values of a passed `raw_transcription_config` are not
+    overwritten by the `None` defaults of the other constructor arguments.
+    """
+
+    raw = aai.RawTranscriptionConfig(
+        language_code="de",
+        punctuate=False,
+        custom_spelling=[{"from": ["k8s"], "to": "Kubernetes"}],
+    )
+
+    config = aai.TranscriptionConfig(raw_transcription_config=raw)
+
+    assert config.raw is raw
+    assert config.language_code == "de"
+    assert config.punctuate is False
+    assert config.custom_spelling == {"Kubernetes": ["k8s"]}
+
+
+def test_explicit_arguments_override_raw_transcription_config():
+    """
+    Tests that explicit constructor arguments take priority over the values of
+    a passed `raw_transcription_config`, and the others are kept.
+    """
+
+    raw = aai.RawTranscriptionConfig(language_code="de", punctuate=False)
+
+    config = aai.TranscriptionConfig(raw_transcription_config=raw, language_code="fr")
+
+    assert config.language_code == "fr"
+    assert config.punctuate is False

tests/unit/test_transcript.py

@@ -577,3 +577,51 @@ def test_metadata_not_present():
     transcript_response = aai.types.TranscriptResponse(**mock_transcript_response)
 
     assert transcript_response.metadata is None
+
+
+def test_config_from_response_with_custom_spelling():
+    """
+    Tests that `transcript.config` rebuilds a config whose response echoes
+    `custom_spelling` in its API shape, a list of from/to objects.
+    """
+
+    mock_transcript_response = factories.generate_dict_factory(
+        factories.TranscriptCompletedResponseFactory
+    )()
+    mock_transcript_response["custom_spelling"] = [
+        {"from": ["k8s", "kube"], "to": "Kubernetes"}
+    ]
+
+    transcript = aai.Transcript.from_response(
+        client=aai.Client.get_default(),
+        response=aai.types.TranscriptResponse(**mock_transcript_response),
+    )
+
+    assert transcript.config.custom_spelling == {"Kubernetes": ["k8s", "kube"]}
+
+
+def test_config_from_response_with_language_detection_results():
+    """
+    Tests that `transcript.config` rebuilds a config whose response carries
+    `language_detection_results`, which is not a `TranscriptionConfig` argument.
+    """
+
+    mock_transcript_response = factories.generate_dict_factory(
+        factories.TranscriptCompletedResponseFactory
+    )()
+    mock_transcript_response["language_detection"] = True
+    mock_transcript_response["language_detection_results"] = {
+        "code_switching_languages": [{"language_code": "es", "confidence": 0.4}]
+    }
+
+    transcript = aai.Transcript.from_response(
+        client=aai.Client.get_default(),
+        response=aai.types.TranscriptResponse(**mock_transcript_response),
+    )
+
+    config = transcript.config
+
+    assert config.language_detection is True
+    results = config.language_detection_results
+    assert results is not None and results.code_switching_languages is not None
+    assert results.code_switching_languages[0].language_code == "es"