fix(azure): address speech transcription review

Co-authored-by: ishaan-berri <ishaan-berri@users.noreply.github.com>

litellm/model_prices_and_context_window_backup.json

@@ -5652,6 +5652,7 @@
         "source": "https://azure.microsoft.com/en-us/pricing/calculator/"
     },
     "azure/speech/azure-stt": {
+        "audio_transcription_config": "azure_speech",
         "input_cost_per_second": 0.0002777778,
         "litellm_provider": "azure",
         "mode": "audio_transcription",

litellm/utils.py

@@ -8060,6 +8060,35 @@ def validate_openai_optional_params(
     return stop
 
 
+@lru_cache(maxsize=1)
+def _get_bundled_model_cost_map() -> Dict[str, Any]:
+    try:
+        model_cost_path = resources.files("litellm").joinpath(
+            "model_prices_and_context_window_backup.json"
+        )
+        return json.loads(model_cost_path.read_text())
+    except Exception:
+        return {}
+
+
+def _get_model_cost_entry_for_provider_config(
+    model: str,
+    provider: LlmProviders,
+) -> Dict[str, Any]:
+    candidate_keys = (model, f"{provider.value}/{model}")
+    for model_key in candidate_keys:
+        model_info = litellm.model_cost.get(model_key)
+        if model_info is not None:
+            return model_info
+
+    bundled_model_cost = _get_bundled_model_cost_map()
+    for model_key in candidate_keys:
+        model_info = bundled_model_cost.get(model_key)
+        if model_info is not None:
+            return model_info
+    return {}
+
+
 class ProviderConfigManager:
     # Dictionary mapping for O(1) provider lookup
     # Stores tuples of (factory_function, needs_model_parameter)
@@ -8500,7 +8529,14 @@ class ProviderConfigManager:
         model: str,
         provider: LlmProviders,
     ) -> Optional[BaseAudioTranscriptionConfig]:
-        if litellm.LlmProviders.AZURE == provider and model.startswith("speech/"):
+        model_cost_entry = _get_model_cost_entry_for_provider_config(
+            model=model,
+            provider=provider,
+        )
+        if (
+            litellm.LlmProviders.AZURE == provider
+            and model_cost_entry.get("audio_transcription_config") == "azure_speech"
+        ):
             from litellm.llms.azure.audio_transcription.transformation import (
                 AzureSpeechAudioTranscriptionConfig,
             )

model_prices_and_context_window.json

@@ -5652,6 +5652,7 @@
         "source": "https://azure.microsoft.com/en-us/pricing/calculator/"
     },
     "azure/speech/azure-stt": {
+        "audio_transcription_config": "azure_speech",
         "input_cost_per_second": 0.0002777778,
         "litellm_provider": "azure",
         "mode": "audio_transcription",

tests/test_litellm/llms/azure/test_azure_speech_audio_transcription.py

@@ -215,3 +215,7 @@ def test_azure_speech_stt_has_non_zero_input_pricing():
     pricing = json.loads(pricing_path.read_text())
 
     assert pricing["azure/speech/azure-stt"]["input_cost_per_second"] > 0
+    assert (
+        pricing["azure/speech/azure-stt"]["audio_transcription_config"]
+        == "azure_speech"
+    )