Remove uv.lock

.gitignore

@@ -51,4 +51,5 @@ docs/api/_build/
 docs/api/api
 
 # uv
-.python-version
+.python-version
+uv.lock

CHANGELOG.md

@@ -18,6 +18,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   async def on_pipeline_finished(task: PipelineTask, frame: Frame):
       ...
   ```
+  
+### Changed
+
+- `torch` and `torchaudio` are no longer required for running Smart Turn
+  locally. This avoids gigabytes of dependencies being installed.
 
 ### Deprecated
 

pyproject.toml

@@ -95,7 +95,7 @@ sambanova = []
 sarvam = [ "websockets>=13.1,<15.0" ]
 sentry = [ "sentry-sdk~=2.23.1" ]
 local-smart-turn = [ "coremltools>=8.0", "transformers", "torch>=2.5.0,<3", "torchaudio>=2.5.0,<3" ]
-local-smart-turn-v3 = [ "transformers", "torch>=2.5.0,<3", "torchaudio>=2.5.0,<3", "onnxruntime>=1.20.1, <2" ]
+local-smart-turn-v3 = [ "transformers", "onnxruntime>=1.20.1, <2" ]
 remote-smart-turn = []
 silero = [ "onnxruntime>=1.20.1, <2" ]
 simli = [ "simli-ai~=0.1.10"]

src/pipecat/audio/turn/smart_turn/local_smart_turn_v3.py

@@ -98,15 +98,15 @@ class LocalSmartTurnAnalyzerV3(BaseSmartTurn):
         inputs = self._feature_extractor(
             audio_array,
             sampling_rate=16000,
-            return_tensors="pt",
+            return_tensors="np",
             padding="max_length",
             max_length=8 * 16000,
             truncation=True,
             do_normalize=True,
         )
 
-        # Convert to numpy and ensure correct shape for ONNX
-        input_features = inputs.input_features.squeeze(0).numpy().astype(np.float32)
+        # Extract features and ensure correct shape for ONNX
+        input_features = inputs.input_features.squeeze(0).astype(np.float32)
         input_features = np.expand_dims(input_features, axis=0)  # Add batch dimension
 
         # Run ONNX inference