feat(py): support numpy inputs

Author: decahedron1

bindings/py/Cargo.toml

@@ -11,3 +11,4 @@ crate-type = ["cdylib"]
 [dependencies]
 earshot = { path = "../.." }
 pyo3 = { version = "0.29", features = [ "extension-module", "abi3-py38" ] }
+numpy = "0.29"

bindings/py/pyproject.toml

@@ -16,6 +16,7 @@ classifiers = [
 	"Topic :: Multimedia :: Sound/Audio :: Analysis",
 	"Topic :: Scientific/Engineering :: Artificial Intelligence"
 ]
+dependencies = ["numpy"]
 
 [project.urls]
 Repository = "https://github.com/pykeio/earshot"

bindings/py/python/earshot/__init__.pyi

@@ -1,5 +1,7 @@
 from typing import Sequence
 
+import numpy
+
 __version__: str
 
 class Detector:
@@ -15,7 +17,7 @@ class Detector:
 		"""
 		...
 
-	def predict_i16(self, frame: Sequence[int]) -> float:
+	def predict_i16(self, frame: Sequence[int] | 'numpy.ndarray') -> float:
 		"""
 		Predicts the voice activity score of a single input frame of 16-bit PCM audio.
 
@@ -28,7 +30,7 @@ class Detector:
 		"""
 		...
 
-	def predict_f32(self, frame: Sequence[float]) -> float:
+	def predict_f32(self, frame: Sequence[float] | 'numpy.ndarray') -> float:
 		"""
 		Predicts the voice activity score of a single input frame of 32-bit floating-point PCM audio.
 

bindings/py/src/lib.rs

@@ -1,4 +1,7 @@
+use std::borrow::Cow;
+
 use earshot::{DefaultPredictor, Detector as RustDetector};
+use numpy::PyArrayLike1;
 use pyo3::{exceptions::PyTypeError, prelude::*};
 
 #[pyclass(module = "earshot._earshot")]
@@ -17,14 +20,24 @@ impl Detector {
 		self.inner.reset();
 	}
 
-	fn predict_i16(&mut self, frame: Vec<i16>) -> PyResult<f32> {
+	fn predict_i16(&mut self, frame: PyArrayLike1<'_, i16>) -> PyResult<f32> {
+		let frame = frame.as_array();
+		let frame = frame
+			.as_slice()
+			.map(Cow::Borrowed)
+			.unwrap_or_else(|| Cow::Owned(frame.iter().copied().collect::<Vec<_>>()));
 		if frame.len() != 256 {
 			return Err(PyErr::new::<PyTypeError, _>("frame must be exactly 256 samples"));
 		}
 		Ok(self.inner.predict_i16(&frame))
 	}
 
-	fn predict_f32(&mut self, frame: Vec<f32>) -> PyResult<f32> {
+	fn predict_f32(&mut self, frame: PyArrayLike1<'_, f32>) -> PyResult<f32> {
+		let frame = frame.as_array();
+		let frame = frame
+			.as_slice()
+			.map(Cow::Borrowed)
+			.unwrap_or_else(|| Cow::Owned(frame.iter().copied().collect::<Vec<_>>()));
 		if frame.len() != 256 {
 			return Err(PyErr::new::<PyTypeError, _>("frame must be exactly 256 samples"));
 		}