[SW] App Lab release 0.3 (arduino#34)

* [app_bricks] Add WaveGenerator brick (arduino#15)

* add external Speaker instance handling

* fix mixer selection in Speaker

* improve volume handling in WaveGenerator and add logging

* add external Speaker instance handling

* fix mixer selection in Speaker

* Enable `cloud_llm` brick (arduino#26)

* Update Cloud LLM brick docs (arduino#27)

* Fix audio clipping and add low-latency WaveGenerator configuration (arduino#30)

* app_bricks/audio-classification: use the brick without init a microphone (arduino#24)

* Fix bricks name model list (arduino#33)

---------

Co-authored-by: Dario Sammaruga <[email protected]>
Co-authored-by: Dario Sammaruga <[email protected]>
Co-authored-by: Stefano Torneo <[email protected]>
Co-authored-by: Stefano Torneo <[email protected]>
Co-authored-by: Ernesto Voltaggio <[email protected]>

Author: 5 people

models/models-list.yaml

@@ -1131,7 +1131,7 @@ models:
       source-model-url: "https://studio.edgeimpulse.com/studio/757509/live"
       private: true
     bricks:
-      - arduino:keyword_spotter
+      - arduino:keyword_spotting
  - updown-wave-motion-detection:
     runner: brick
     name : "Continuous motion detection"
@@ -1182,4 +1182,4 @@ models:
       source-model-url: "https://studio.edgeimpulse.com/public/749446/live"
       private: true
     bricks:
-      - arduino:audio_classifier
+      - arduino:audio_classification

src/arduino/app_bricks/audio_classification/__init__.py

@@ -66,7 +66,8 @@ def stop(self):
         """
         super().stop()
 
-    def classify_from_file(self, audio_path: str, confidence: int = None) -> dict | None:
+    @staticmethod
+    def classify_from_file(audio_path: str, confidence: float = 0.8) -> dict | None:
         """Classify audio content from a WAV file.
 
         Supported sample widths:
@@ -77,9 +78,8 @@ def classify_from_file(self, audio_path: str, confidence: int = None) -> dict |
 
         Args:
             audio_path (str): Path to the `.wav` audio file to classify.
-            confidence (int, optional): Confidence threshold (0–1). If None,
-                the default confidence level specified during initialization
-                will be applied.
+            confidence (float, optional): Minimum confidence threshold (0.0–1.0) required
+                for a detection to be considered valid. Defaults to 0.8 (80%).
 
         Returns:
             dict | None: A dictionary with keys:
@@ -121,9 +121,8 @@ def classify_from_file(self, audio_path: str, confidence: int = None) -> dict |
                     features = list(struct.unpack(fmt, frames))
                 else:
                     raise ValueError(f"Unsupported sample width: {samp_width} bytes. Cannot process this WAV file.")
-
-                classification = super().infer_from_features(features[: int(self.model_info.input_features_count)])
-                best_match = super().get_best_match(classification, confidence)
+                classification = AudioClassification.infer_from_features(features)
+                best_match = AudioDetector.get_best_match(classification, confidence)
                 if not best_match:
                     return None
                 keyword, confidence = best_match

src/arduino/app_bricks/audio_classification/examples/2_glass_breaking_from_file.py

@@ -6,7 +6,5 @@
 # EXAMPLE_REQUIRES = "Requires an audio file with the glass breaking sound."
 from arduino.app_bricks.audio_classification import AudioClassification
 
-classifier = AudioClassification()
-
-classification = classifier.classify_from_file("glass_breaking.wav")
+classification = AudioClassification.classify_from_file("glass_breaking.wav")
 print("Result:", classification)

src/arduino/app_bricks/motion_detection/__init__.py

@@ -133,7 +133,7 @@ def _detection_loop(self):
             return
 
         try:
-            ret = super().infer_from_features(features[: int(self._model_info.input_features_count)].flatten().tolist())
+            ret = self.infer_from_features(features.flatten().tolist())
             spotted_movement = self._movement_spotted(ret)
             if spotted_movement is not None:
                 keyword, confidence, complete_detection = spotted_movement

src/arduino/app_bricks/vibration_anomaly_detection/__init__.py

@@ -133,7 +133,7 @@ def loop(self):
             if features is None or len(features) == 0:
                 return
 
-            ret = super().infer_from_features(features[: int(self._model_info.input_features_count)].flatten().tolist())
+            ret = self.infer_from_features(features.flatten().tolist())
             logger.debug(f"Inference result: {ret}")
             spotted_anomaly = self._extract_anomaly_score(ret)
             if spotted_anomaly is not None:

src/arduino/app_internal/core/audio.py

@@ -85,17 +85,24 @@ def stop(self):
             self._mic.stop()
         self._buffer.flush()
 
-    def get_best_match(self, item: dict, confidence: int = None) -> tuple[str, float] | None:
+    @staticmethod
+    def get_best_match(item: dict, confidence: float) -> tuple[str, float] | None:
         """Extract the best matched keyword from the classification results.
 
         Args:
         item (dict): The classification result from the inference.
-        confidence (int): The confidence threshold for classification. If None, uses the instance's confidence level.
+        confidence (float): The confidence threshold for classification.
 
         Returns:
         tuple[str, float] | None: The best matched keyword and its confidence, or None if no match is found.
+
+        Raises:
+        ValueError: If confidence level is not provided.
         """
-        classification = _extract_classification(item, confidence or self.confidence)
+        if confidence is None:
+            raise ValueError("Confidence level must be provided.")
+
+        classification = _extract_classification(item, confidence)
         if not classification:
             return None
 
@@ -141,8 +148,8 @@ def _inference_loop(self):
 
         logger.debug(f"Processing sensor data with {len(features)} features.")
         try:
-            ret = super().infer_from_features(features[: int(self.model_info.input_features_count)].tolist())
-            spotted_keyword = self.get_best_match(ret)
+            ret = self.infer_from_features(features.tolist())
+            spotted_keyword = self.get_best_match(ret, self.confidence)
             if spotted_keyword:
                 keyword, confidence = spotted_keyword
                 keyword = keyword.lower()

src/arduino/app_internal/core/ei.py

@@ -42,18 +42,13 @@ class EdgeImpulseRunnerFacade:
     """Facade for Edge Impulse Object Detection and Classification."""
 
     def __init__(self):
-        """Initialize the EdgeImpulseRunnerFacade with the API path."""
-        infra = load_brick_compose_file(self.__class__)
-        for k, v in infra["services"].items():
-            self.host = k
-            self.infra = v
-            break  # Only one service is expected
-
-        self.host = resolve_address(self.host)
+        """Initialize the EdgeImpulseRunnerFacade with the API path.
 
-        self.port = 1337  # Default EI HTTP port
-        self.url = f"http://{self.host}:{self.port}"
-        logger.warning(f"[{self.__class__.__name__}] Host: {self.host} - Ports: {self.port} - URL: {self.url}")
+        Raises:
+            RuntimeError: If the Edge Impulse runner address cannot be resolved.
+        """
+        self.url = self._get_ei_url()
+        logger.info(f"[{self.__class__.__name__}] URL: {self.url}")
 
     def infer_from_file(self, image_path: str) -> dict | None:
         if not image_path or image_path == "":
@@ -124,47 +119,58 @@ def process(self, item):
             logger.error(f"[{self.__class__}] Error processing file {item}: {e}")
         return None
 
-    def infer_from_features(self, features: list) -> dict | None:
-        """Infer from features using the Edge Impulse API.
+    @classmethod
+    def infer_from_features(cls, features: list) -> dict | None:
+        """
+        Infer from features using the Edge Impulse API.
 
         Args:
+            cls: The class method caller.
             features (list): A list of features to send to the Edge Impulse API.
 
         Returns:
             dict | None: The response from the Edge Impulse API as a dictionary, or None if an error occurs.
         """
         try:
-            response = requests.post(f"{self.url}/api/features", json={"features": features})
+            url = cls._get_ei_url()
+            model_info = cls.get_model_info(url)
+            features = features[: int(model_info.input_features_count)]
+
+            response = requests.post(f"{url}/api/features", json={"features": features})
             if response.status_code == 200:
                 return response.json()
             else:
-                logger.warning(f"[{self.__class__}] error: {response.status_code}. Message: {response.text}")
+                logger.warning(f"[{cls.__name__}] error: {response.status_code}. Message: {response.text}")
                 return None
         except Exception as e:
-            logger.error(f"[{self.__class__.__name__}] Error: {e}")
+            logger.error(f"[{cls.__name__}] Error: {e}")
             return None
 
-    def get_model_info(self) -> EdgeImpulseModelInfo | None:
+    @classmethod
+    def get_model_info(cls, url: str = None) -> EdgeImpulseModelInfo | None:
         """Get model information from the Edge Impulse API.
 
+        Args:
+            cls: The class method caller.
+            url (str): The base URL of the Edge Impulse API. If None, it will be determined automatically.
+
         Returns:
             model_info (EdgeImpulseModelInfo | None): An instance of EdgeImpulseModelInfo containing model details, None if an error occurs.
         """
-        if not self.host or not self.port:
-            logger.error(f"[{self.__class__}] Host or port not set. Cannot fetch model info.")
-            return None
+        if not url:
+            url = cls._get_ei_url()
 
         http_client = HttpClient(total_retries=6)  # Initialize the HTTP client with retry logic
         try:
-            response = http_client.request_with_retry(f"{self.url}/api/info")
+            response = http_client.request_with_retry(f"{url}/api/info")
             if response.status_code == 200:
-                logger.debug(f"[{self.__class__.__name__}] Fetching model info from {self.url}/api/info -> {response.status_code} {response.json}")
+                logger.debug(f"[{cls.__name__}] Fetching model info from {url}/api/info -> {response.status_code} {response.json}")
                 return EdgeImpulseModelInfo(response.json())
             else:
-                logger.warning(f"[{self.__class__}] Error fetching model info: {response.status_code}. Message: {response.text}")
+                logger.warning(f"[{cls}] Error fetching model info: {response.status_code}. Message: {response.text}")
                 return None
         except Exception as e:
-            logger.error(f"[{self.__class__}] Error fetching model info: {e}")
+            logger.error(f"[{cls}] Error fetching model info: {e}")
             return None
         finally:
             http_client.close()  # Close the HTTP client session
@@ -237,3 +243,19 @@ def _extract_anomaly_score(self, item: dict):
                 return class_results["anomaly"]
 
         return None
+
+    @classmethod
+    def _get_ei_url(cls):
+        infra = load_brick_compose_file(cls)
+        if not infra or "services" not in infra:
+            raise RuntimeError("Cannot load Brick Compose file to resolve Edge Impulse runner address.")
+        host = None
+        for k, v in infra["services"].items():
+            host = k
+            break
+        if not host:
+            raise RuntimeError("Cannot resolve Edge Impulse runner address from Brick Compose file.")
+        addr = resolve_address(host)
+        if not addr:
+            raise RuntimeError("Host address resolution failed for Edge Impulse runner.")
+        return f"http://{addr}:1337"

tests/arduino/app_bricks/motion_detection/test_motion_detection.py

@@ -24,10 +24,10 @@ def app_instance(monkeypatch):
 @pytest.fixture(autouse=True)
 def mock_dependencies(monkeypatch: pytest.MonkeyPatch):
     """Mock out docker-compose lookups and image helpers."""
-    fake_compose = {"services": {"models-runner": {"ports": ["${BIND_ADDRESS:-127.0.0.1}:${BIND_PORT:-8100}:8100"]}}}
-    monkeypatch.setattr("arduino.app_internal.core.load_brick_compose_file", lambda cls: fake_compose)
+    fake_compose = {"services": {"ei-inference": {"ports": ["${BIND_ADDRESS:-127.0.0.1}:${BIND_PORT:-1337}:1337"]}}}
+    monkeypatch.setattr("arduino.app_internal.core.ei.load_brick_compose_file", lambda cls: fake_compose)
     monkeypatch.setattr("arduino.app_internal.core.resolve_address", lambda host: "127.0.0.1")
-    monkeypatch.setattr("arduino.app_internal.core.parse_docker_compose_variable", lambda x: [(None, None), (None, "8200")])
+    monkeypatch.setattr("arduino.app_internal.core.parse_docker_compose_variable", lambda x: [(None, None), (None, "1337")])
 
     class FakeResp:
         status_code = 200

tests/arduino/app_bricks/objectdetection/test_objectdetection.py

@@ -7,6 +7,7 @@
 import io
 from PIL import Image
 from arduino.app_bricks.object_detection import ObjectDetection
+from arduino.app_utils import HttpClient
 
 
 class ModelInfo:
@@ -20,12 +21,65 @@ def mock_dependencies(monkeypatch: pytest.MonkeyPatch):
 
     This is needed to avoid network calls and other side effects.
     """
-    fake_compose = {"services": {"models-runner": {"ports": ["${BIND_ADDRESS:-127.0.0.1}:${BIND_PORT:-8100}:8100"]}}}
-    monkeypatch.setattr("arduino.app_internal.core.load_brick_compose_file", lambda cls: fake_compose)
+    fake_compose = {"services": {"ei-inference": {"ports": ["${BIND_ADDRESS:-127.0.0.1}:${BIND_PORT:-1337}:1337"]}}}
+    monkeypatch.setattr("arduino.app_internal.core.ei.load_brick_compose_file", lambda cls: fake_compose)
     monkeypatch.setattr("arduino.app_internal.core.resolve_address", lambda host: "127.0.0.1")
-    monkeypatch.setattr("arduino.app_internal.core.parse_docker_compose_variable", lambda x: [(None, None), (None, "8100")])
+    monkeypatch.setattr("arduino.app_internal.core.parse_docker_compose_variable", lambda x: [(None, None), (None, "1337")])
     monkeypatch.setattr("arduino.app_bricks.object_detection.ObjectDetection.get_model_info", lambda self: ModelInfo("object-detection"))
 
+    class FakeResp:
+        status_code = 200
+
+        def json(self):
+            return {
+                "project": {
+                    "deploy_version": 11,
+                    "id": 774707,
+                    "impulse_id": 1,
+                    "impulse_name": "Time series data, Spectral Analysis, Classification (Keras), Anomaly Detection (K-means)",
+                    "name": "Fan Monitoring - Advanced Anomaly Detection",
+                    "owner": "Arduino",
+                },
+                "modelParameters": {
+                    "has_visual_anomaly_detection": False,
+                    "axis_count": 3,
+                    "frequency": 100,
+                    "has_anomaly": 1,
+                    "has_object_tracking": False,
+                    "has_performance_calibration": False,
+                    "image_channel_count": 0,
+                    "image_input_frames": 0,
+                    "image_input_height": 0,
+                    "image_input_width": 0,
+                    "image_resize_mode": "none",
+                    "inferencing_engine": 4,
+                    "input_features_count": 600,
+                    "interval_ms": 10,
+                    "label_count": 2,
+                    "labels": ["nominal", "off"],
+                    "model_type": "classification",
+                    "sensor": 2,
+                    "slice_size": 50,
+                    "thresholds": [],
+                    "use_continuous_mode": False,
+                    "sensorType": "accelerometer",
+                },
+            }
+
+    def fake_get(
+        self,
+        url: str,
+        method: str = "GET",
+        data: dict | str = None,
+        json: dict = None,
+        headers: dict = None,
+        timeout: int = 5,
+    ):
+        return FakeResp()
+
+    # Mock the requests.get method to return a fake response
+    monkeypatch.setattr(HttpClient, "request_with_retry", fake_get)
+
 
 @pytest.fixture
 def detector():

tests/arduino/app_bricks/vibration_anomaly_detection/test_vibration_anomaly_detection.py

@@ -24,10 +24,10 @@ def app_instance(monkeypatch):
 @pytest.fixture(autouse=True)
 def mock_dependencies(monkeypatch: pytest.MonkeyPatch):
     """Mock out docker-compose lookups and image helpers."""
-    fake_compose = {"services": {"models-runner": {"ports": ["${BIND_ADDRESS:-127.0.0.1}:${BIND_PORT:-8100}:8100"]}}}
-    monkeypatch.setattr("arduino.app_internal.core.load_brick_compose_file", lambda cls: fake_compose)
+    fake_compose = {"services": {"ei-inference": {"ports": ["${BIND_ADDRESS:-127.0.0.1}:${BIND_PORT:-1337}:1337"]}}}
+    monkeypatch.setattr("arduino.app_internal.core.ei.load_brick_compose_file", lambda cls: fake_compose)
     monkeypatch.setattr("arduino.app_internal.core.resolve_address", lambda host: "127.0.0.1")
-    monkeypatch.setattr("arduino.app_internal.core.parse_docker_compose_variable", lambda x: [(None, None), (None, "8200")])
+    monkeypatch.setattr("arduino.app_internal.core.parse_docker_compose_variable", lambda x: [(None, None), (None, "1337")])
 
     class FakeResp:
         status_code = 200