jaxl-python 0.0.35__tar.gz → 0.0.36__tar.gz

{jaxl_python-0.0.35 → jaxl_python-0.0.36}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: jaxl-python
-Version: 0.0.35
+Version: 0.0.36
 Summary: Official Python SDK and CLI for interfacing with the Jaxl API. Manage calls, messages, IVRs, devices, teams, payments, campaigns, streaming transcriptions and AI Agents. Built on OpenAPI, it offers both generated API clients and Pythonic wrappers for ease of use.
 Home-page: https://github.com/jaxl-innovations-private-limited/jaxl-python
 Download-URL: https://github.com/jaxl-innovations-private-limited/jaxl-python/archive/main.zip

{jaxl_python-0.0.35 → jaxl_python-0.0.36}/jaxl/api/client/_scm_version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.0.35'
-__version_tuple__ = version_tuple = (0, 0, 35)
+__version__ = version = '0.0.36'
+__version_tuple__ = version_tuple = (0, 0, 36)

{jaxl_python-0.0.35 → jaxl_python-0.0.36}/jaxl/api/resources/apps.py

@@ -67,8 +67,8 @@ def _start_server(
     transcribe_device: str = "cpu",
     transcribe_temperature: float = 0.3,
     vad_aggressiveness: int = 2,
-    vad_silence_frame_threshold: int = 25,
-    vad_speech_frame_threshold: int = 20,
+    vad_silence_frame_threshold: int = 12,
+    vad_speech_frame_threshold: int = 8,
 ) -> "FastAPI":
     from fastapi import FastAPI, Request, WebSocket
 
@@ -239,7 +239,7 @@ def _start_server(
             speech_frame_threshold=vad_speech_frame_threshold,
         )
         speaking: bool = False
-        buffer: Deque[bytes] = deque(maxlen=sdetector.speech_frame_threshold)
+        buffer: Deque[bytes] = deque(maxlen=4)
         slin16s: List[bytes] = []
 
         await ws.accept()
@@ -259,25 +259,19 @@ def _start_server(
                     # Detect start/end of speech
                     buffer.append(slin16)
                     change = sdetector.process(slin16)
+                    current_frame_is_speech = sdetector.last_frame_is_speech
                     # Manage speech segments
                     if change is True:
                         speaking = change
                         await app.handle_speech_detection(state["call_id"], speaking)
                         if len(slin16s) == 0:
-                            # Silence just got detected, copy over
-                            # last speech_frame_threshold of frames
+                            # Copy over a short, fixed preroll window.
                             slin16s = list(buffer)
                             if len(slin16s) > 0:
                                 await app.handle_speech_chunks(req, slin16s)
-                            # print("💿")
-                        # print("🎙️")
-                        slin16s.append(slin16)
-                        await app.handle_speech_chunks(req, [slin16])
                     elif change is False:
                         speaking = change
-                        await app.handle_speech_chunks(req, [slin16])
                         await app.handle_speech_detection(state["call_id"], speaking)
-                        # print("🤐")
                         if len(slin16s) > 0:
                             # Invoke speech segment handlers
                             await app.handle_speech_segment(req, slin16s)
@@ -296,8 +290,9 @@ def _start_server(
                     else:
                         assert change is None
                         if speaking is True:
-                            await app.handle_speech_chunks(req, [slin16])
-                            slin16s.append(slin16)
+                            if current_frame_is_speech:
+                                await app.handle_speech_chunks(req, [slin16])
+                                slin16s.append(slin16)
                         else:
                             assert speaking is False
                 elif ev == "connected":
@@ -426,20 +421,20 @@ def _subparser(parser: argparse.ArgumentParser) -> None:
     apps_run_parser.add_argument(
         "--vad-aggressiveness",
         type=int,
-        default=1,
+        default=2,
         help="VAD Aggressiveness",
     )
     apps_run_parser.add_argument(
         "--vad-silence-frame-threshold",
         type=int,
-        default=25,
-        help="Amount of silence in milliseconds before silence detection is triggered",
+        default=12,
+        help="Number of 20ms frames of silence before silence detection is triggered",
     )
     apps_run_parser.add_argument(
         "--vad-speech-frame-threshold",
         type=int,
-        default=20,
-        help="Amount of speech in milliseconds before speech detection is triggered",
+        default=8,
+        help="Number of 20ms frames of speech before speech detection is triggered",
     )
     apps_run_parser.set_defaults(
         func=apps_run,

{jaxl_python-0.0.35 → jaxl_python-0.0.36}/jaxl/api/resources/silence.py

@@ -20,8 +20,8 @@ class SilenceDetector:
         sample_rate: int = 8000,
         frame_duration_ms: int = 20,
         aggressiveness: int = 2,
-        silence_frame_threshold: int = 25,  # ~500ms
-        speech_frame_threshold: int = 20,  # ~400ms
+        silence_frame_threshold: int = 12,  # ~240ms
+        speech_frame_threshold: int = 8,  # ~160ms
     ):
         import webrtcvad
 
@@ -37,6 +37,7 @@ class SilenceDetector:
         self.speech_frames = 0
         self.silence_frames = 0
         self.buffer = b""
+        self.last_frame_is_speech: Optional[bool] = None
 
     def process(self, slin16: bytes) -> Optional[bool]:
         """Process PCM16 mono audio. Returns:
@@ -54,6 +55,7 @@ class SilenceDetector:
             )
 
             is_speech = self.vad.is_speech(frame, self.sample_rate)
+            self.last_frame_is_speech = is_speech
 
             if is_speech:
                 self.speech_frames += 1
@@ -81,3 +83,4 @@ class SilenceDetector:
         self.silence_frames = 0
         self.speech_frames = 0
         self.is_talking = False
+        self.last_frame_is_speech = None

{jaxl_python-0.0.35 → jaxl_python-0.0.36}/jaxl_python.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: jaxl-python
-Version: 0.0.35
+Version: 0.0.36
 Summary: Official Python SDK and CLI for interfacing with the Jaxl API. Manage calls, messages, IVRs, devices, teams, payments, campaigns, streaming transcriptions and AI Agents. Built on OpenAPI, it offers both generated API clients and Pythonic wrappers for ease of use.
 Home-page: https://github.com/jaxl-innovations-private-limited/jaxl-python
 Download-URL: https://github.com/jaxl-innovations-private-limited/jaxl-python/archive/main.zip