python-aidot-cameras 0.7.1__tar.gz → 0.7.3__tar.gz

{python_aidot_cameras-0.7.1/src/python_aidot_cameras.egg-info → python_aidot_cameras-0.7.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-aidot-cameras
-Version: 0.7.1
+Version: 0.7.3
 Summary: Control AiDot/Leedarson WiFi lights and cameras (WebRTC streaming, two-way audio, PTZ, controls)
 Author-email: cbrightly <chris.brightly@gmail.com>
 License-Expression: MIT

{python_aidot_cameras-0.7.1 → python_aidot_cameras-0.7.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "python-aidot-cameras"
-version = "0.7.1"
+version = "0.7.3"
 description = "Control AiDot/Leedarson WiFi lights and cameras (WebRTC streaming, two-way audio, PTZ, controls)"
 readme = "README.md"
 license = "MIT"

{python_aidot_cameras-0.7.1 → python_aidot_cameras-0.7.3}/src/aidot/camera/client.py

@@ -2168,8 +2168,15 @@ def _dtls_av_mux_run(vq, aq, out_fileobj, progress, stop_flag) -> None:
             _target = _db2amp(float(os.environ.get("AIDOT_AUDIO_TARGET_DBFS", "-15"))) * 32767.0
             _maxg = _db2amp(float(os.environ.get("AIDOT_AUDIO_MAXGAIN_DB", "30")))
             _ming = _db2amp(float(os.environ.get("AIDOT_AUDIO_MINGAIN_DB", "-12")))
+            # Noise gate: below this RMS the input is (near) silence, so the AGC
+            # must NOT crank toward +maxg - doing so amplifies the A-law
+            # quantization floor into audible high-frequency clicking on a quiet
+            # camera.  Gain is scaled down quadratically below the gate.  Set very
+            # low (e.g. -120) to disable.
+            _gate = _db2amp(float(os.environ.get("AIDOT_AUDIO_GATE_DBFS", "-45"))) * 32767.0
         except (ValueError, TypeError):
             _target, _maxg, _ming = _db2amp(-15) * 32767.0, _db2amp(30), _db2amp(-12)
+            _gate = _db2amp(-45) * 32767.0
         _agc_ms = [None]   # smoothed mean-square (level tracker)
         try:
             import numpy as _np
@@ -2204,7 +2211,7 @@ def _dtls_av_mux_run(vq, aq, out_fileobj, progress, stop_flag) -> None:
             except Exception:
                 pass
 
-    def _flush_audio():
+    def _flush_audio(drain=False):
         if not have_audio:
             try:
                 while True:
@@ -2232,6 +2239,12 @@ def _dtls_av_mux_run(vq, aq, out_fileobj, progress, stop_flag) -> None:
                                 _agc_ms[0] = _agc_ms[0] * 0.95 + _ms * 0.05
                             _rms = (_agc_ms[0] ** 0.5) + 1.0
                             _gain = min(_maxg, max(_ming, _target / _rms))
+                            # Noise gate: when the level is below the gate floor the
+                            # signal is essentially silence; scale the gain down
+                            # quadratically so the A-law quantization noise is not
+                            # amplified into audible clicking.
+                            if _rms < _gate:
+                                _gain *= (_rms / _gate) ** 2
                             # tanh soft-limiter catches transient peaks past target.
                             _y = _np.tanh(_x * (_gain / 32767.0)) * 32767.0
                             _g = av.AudioFrame.from_ndarray(
@@ -2247,10 +2260,16 @@ def _dtls_av_mux_run(vq, aq, out_fileobj, progress, stop_flag) -> None:
         while True:
             fr = fifo.read(1024)   # AAC wants 1024-sample frames
             if fr is None:
-                # 160 PCMA samples → 960 resampled @ 48 kHz, but AAC needs 1024.
-                # Every ~17 packets (~340 ms) the FIFO falls 64 samples short
-                # and would skip an entire 21 ms frame - audible choppiness.
-                # Pad the leftover with silence so the frame is complete.
+                # FIFO has < 1024 samples.  The audio source is continuous
+                # (measured: 100% timeline coverage, no lost packets), so the
+                # remainder must stay in the FIFO and be completed by the next
+                # decoded packet - NOT padded with silence here.  Padding on
+                # every 5 ms flush injected a micro-silence whenever the FIFO was
+                # momentarily short (almost always, given ~40 ms packet spacing) -
+                # that was the choppiness.  Only on the final drain do we pad the
+                # last partial frame so the encoder can flush it.
+                if not drain:
+                    break
                 s = getattr(fifo, "samples", 0)
                 if s == 0:
                     break
@@ -2288,7 +2307,7 @@ def _dtls_av_mux_run(vq, aq, out_fileobj, progress, stop_flag) -> None:
         _flush_audio()
         _t.sleep(0.005)  # 5 ms - processes audio within 5 ms of arrival vs 20 ms
     _flush_video()
-    _flush_audio()
+    _flush_audio(drain=True)  # final: pad the last partial frame so it flushes
     try:
         if have_audio:
             for opkt in aenc.encode(None):  # flush

{python_aidot_cameras-0.7.1 → python_aidot_cameras-0.7.3/src/python_aidot_cameras.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-aidot-cameras
-Version: 0.7.1
+Version: 0.7.3
 Summary: Control AiDot/Leedarson WiFi lights and cameras (WebRTC streaming, two-way audio, PTZ, controls)
 Author-email: cbrightly <chris.brightly@gmail.com>
 License-Expression: MIT