From d9c9aefdb385cb7db8730e64be3c8e3b95d560d7 Mon Sep 17 00:00:00 2001
From: Dmitry Dementev <me@dementev.space>
Date: Sat, 25 Apr 2026 21:22:02 +0300
Subject: [PATCH] =?UTF-8?q?feat(onnx-asr):=20=D1=80=D0=B5=D0=B0=D0=BB?=
 =?UTF-8?q?=D0=B8=D0=B7=D0=BE=D0=B2=D0=B0=D1=82=D1=8C=20create=5Fmodel=20?=
 =?UTF-8?q?=D1=81=20Silero=20VAD?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Зачем:
  - onnx-asr модель должна создаваться с квантизацией и VAD для разбивки аудио на сегменты.
- Что:
  - метод create_model вызывает onnx_asr.load_model с квантизацией и cpu_preprocessing=True.
  - подгружается Silero VAD, прикрепляется к модели через with_vad.
  - написаны 3 теста: проверка аргументов load_model, загрузка VAD, передача fp16.
- Проверка:
  - uv run pytest tests/test_onnx_asr.py -v
---
 src/local_transcriber/backends/onnx_asr.py | 26 ++++++++
 tests/test_onnx_asr.py                     | 73 ++++++++++++++++++++++
 2 files changed, 99 insertions(+)

diff --git a/src/local_transcriber/backends/onnx_asr.py b/src/local_transcriber/backends/onnx_asr.py
index 770c412..7411767 100644
--- a/src/local_transcriber/backends/onnx_asr.py
+++ b/src/local_transcriber/backends/onnx_asr.py
@@ -36,6 +36,32 @@ class OnnxAsrBackend:
         self._resolved_model_id = self._resolve_model(model_name)
         return self._resolved_model_id
 
+    def create_model(
+        self,
+        model_path: str,
+        device: str,
+        compute_type: str,
+        cpu_threads: int = 0,
+    ) -> Any:
+        """Creates onnx-asr model with VAD.
+
+        model_path: onnx-asr model identifier (e.g. "gigaam-v3-ctc").
+        compute_type: "int8", "fp16", or "float32" — passed as quantization.
+        cpu_threads: not used by onnx-asr (onnxruntime manages threads internally).
+        """
+        import onnx_asr
+
+        ct = compute_type if compute_type in ("int8", "fp16", "float32") else "int8"
+
+        model = onnx_asr.load_model(
+            model=model_path,
+            quantization=ct,
+            cpu_preprocessing=True,
+        )
+        vad = onnx_asr.load_vad("silero")
+        self._vad = vad
+        return model.with_vad(vad)
+
     def _resolve_model(self, model_name: str) -> str:
         """Resolve alias to onnx-asr model name. Raw names pass through."""
         if model_name in MODEL_ALIASES:
diff --git a/tests/test_onnx_asr.py b/tests/test_onnx_asr.py
index a1f842d..7a6a81b 100644
--- a/tests/test_onnx_asr.py
+++ b/tests/test_onnx_asr.py
@@ -22,6 +22,79 @@ class TestEnsureModelAvailable:
         assert backend.actual_compute_type == "float32"
 
 
+class TestCreateModel:
+    def test_calls_load_model_with_correct_args(self, monkeypatch):
+        """Verify create_model passes correct args to onnx_asr.load_model."""
+        calls = []
+
+        def fake_load_model(model=None, path=None, quantization=None,
+                           cpu_preprocessing=None, **kwargs):
+            calls.append({
+                "model": model, "path": path, "quantization": quantization,
+                "cpu_preprocessing": cpu_preprocessing,
+            })
+            return FakeAsrAdapter()
+
+        class FakeAsrAdapter:
+            def with_vad(self, vad):
+                return self
+
+        monkeypatch.setattr("onnx_asr.load_model", fake_load_model)
+
+        backend = OnnxAsrBackend()
+        backend.actual_compute_type = "int8"
+        model = backend.create_model("gigaam-v3-ctc", "onnx", "int8")
+
+        assert len(calls) == 1
+        assert calls[0]["quantization"] == "int8"
+        assert calls[0]["cpu_preprocessing"] is True
+        assert model is not None
+
+    def test_loads_silero_vad(self, monkeypatch):
+        """Verify Silero VAD is loaded and attached to model."""
+        vad_calls = []
+
+        def fake_load_vad(model, **kwargs):
+            vad_calls.append(model)
+            return "fake_vad"
+
+        def fake_load_model(**kwargs):
+            return FakeAsrAdapter()
+
+        class FakeAsrAdapter:
+            def with_vad(self, vad):
+                self._vad = vad
+                return self
+
+        monkeypatch.setattr("onnx_asr.load_model", fake_load_model)
+        monkeypatch.setattr("onnx_asr.load_vad", fake_load_vad)
+
+        backend = OnnxAsrBackend()
+        model = backend.create_model("gigaam-v3-ctc", "onnx", "int8")
+
+        assert vad_calls == ["silero"]
+
+    def test_fp16_compute_type(self, monkeypatch):
+        """Verify fp16 compute_type is passed through."""
+        calls = []
+
+        def fake_load_model(model=None, quantization=None, **kwargs):
+            calls.append(quantization)
+            return FakeAsrAdapter()
+
+        class FakeAsrAdapter:
+            def with_vad(self, vad):
+                return self
+
+        monkeypatch.setattr("onnx_asr.load_model", fake_load_model)
+        monkeypatch.setattr("onnx_asr.load_vad", lambda model, **kw: None)
+
+        backend = OnnxAsrBackend()
+        backend.create_model("parakeet-v3", "onnx", "fp16")
+
+        assert calls == ["fp16"]
+
+
 class TestModelAliases:
     def test_gigaam_v3_resolves(self):
         backend = OnnxAsrBackend()