From a1d2f98765de5e0d852adb13dbc1d8f189a458e6 Mon Sep 17 00:00:00 2001
From: Nick Mowen <nickmowen213@gmail.com>
Date: Sun, 16 Jul 2023 16:54:23 -0600
Subject: [PATCH] Add audio filters config

---
 frigate/config.py       | 18 +++++++++++++++++-
 frigate/const.py        |  1 +
 frigate/events/audio.py |  6 ++++--
 3 files changed, 22 insertions(+), 3 deletions(-)

diff --git a/frigate/config.py b/frigate/config.py
index bca0ba080..59e086989 100644
--- a/frigate/config.py
+++ b/frigate/config.py
@@ -11,7 +11,13 @@ import numpy as np
 from pydantic import BaseModel, Extra, Field, parse_obj_as, validator
 from pydantic.fields import PrivateAttr
 
-from frigate.const import CACHE_DIR, DEFAULT_DB_PATH, REGEX_CAMERA_NAME, YAML_EXT
+from frigate.const import (
+    AUDIO_MIN_CONFIDENCE,
+    CACHE_DIR,
+    DEFAULT_DB_PATH,
+    REGEX_CAMERA_NAME,
+    YAML_EXT,
+)
 from frigate.detectors import DetectorConfig, ModelConfig
 from frigate.detectors.detector_config import BaseDetectorConfig
 from frigate.ffmpeg_presets import (
@@ -334,6 +340,15 @@ class FilterConfig(FrigateBaseModel):
     )
 
 
+class AudioFilterConfig(FrigateBaseModel):
+    threshold: float = Field(
+        default=0.8,
+        ge=AUDIO_MIN_CONFIDENCE,
+        lt=1.0,
+        title="Minimum detection confidence threshold for audio to be counted.",
+    )
+
+
 class RuntimeFilterConfig(FilterConfig):
     mask: Optional[np.ndarray]
     raw_mask: Optional[Union[str, List[str]]]
@@ -424,6 +439,7 @@ class AudioConfig(FrigateBaseModel):
     listen: List[str] = Field(
         default=DEFAULT_LISTEN_AUDIO, title="Audio to listen for."
     )
+    filters: Optional[Dict[str, AudioFilterConfig]] = Field(title="Audio filters.")
     enabled_in_config: Optional[bool] = Field(
         title="Keep track of original state of audio detection."
     )
diff --git a/frigate/const.py b/frigate/const.py
index b6b0e44bd..4dec7f366 100644
--- a/frigate/const.py
+++ b/frigate/const.py
@@ -28,6 +28,7 @@ AUDIO_DURATION = 0.975
 AUDIO_FORMAT = "s16le"
 AUDIO_MAX_BIT_RANGE = 32768.0
 AUDIO_SAMPLE_RATE = 16000
+AUDIO_MIN_CONFIDENCE = 0.5
 
 # Regex Consts
 
diff --git a/frigate/events/audio.py b/frigate/events/audio.py
index 446920c89..abd491ad7 100644
--- a/frigate/events/audio.py
+++ b/frigate/events/audio.py
@@ -19,6 +19,7 @@ from frigate.const import (
     AUDIO_DURATION,
     AUDIO_FORMAT,
     AUDIO_MAX_BIT_RANGE,
+    AUDIO_MIN_CONFIDENCE,
     AUDIO_SAMPLE_RATE,
     CACHE_DIR,
     FRIGATE_LOCALHOST,
@@ -130,7 +131,7 @@ class AudioTfl:
 
         return detections
 
-    def detect(self, tensor_input, threshold=0.8):
+    def detect(self, tensor_input, threshold=AUDIO_MIN_CONFIDENCE):
         detections = []
 
         if self.stop_event.is_set():
@@ -200,7 +201,8 @@ class AudioEventMaintainer(threading.Thread):
                 if label not in self.config.audio.listen:
                     continue
 
-                self.handle_detection(label, score)
+                if score > self.config.audio.filters.get(label, {}).get("threshold", 0):
+                    self.handle_detection(label, score)
 
         self.expire_detections()