jaxl-python 0.0.37__tar.gz → 0.0.38__tar.gz

{jaxl_python-0.0.37 → jaxl_python-0.0.38}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: jaxl-python
-Version: 0.0.37
+Version: 0.0.38
 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-0.0.37 → jaxl_python-0.0.38}/jaxl/api/client/_scm_version.py

@@ -17,5 +17,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.0.37'
-__version_tuple__ = version_tuple = (0, 0, 37)
+__version__ = version = '0.0.38'
+__version_tuple__ = version_tuple = (0, 0, 38)

{jaxl_python-0.0.37 → jaxl_python-0.0.38}/jaxl/api/resources/apps.py

@@ -68,7 +68,7 @@ def _start_server(
     transcribe_temperature: float = 0.3,
     vad_aggressiveness: int = 2,
     vad_silence_frame_threshold: int = 12,
-    vad_speech_frame_threshold: int = 8,
+    vad_speech_frame_threshold: int = 5,
 ) -> "FastAPI":
     from fastapi import FastAPI, Request, WebSocket
 
@@ -433,7 +433,7 @@ def _subparser(parser: argparse.ArgumentParser) -> None:
     apps_run_parser.add_argument(
         "--vad-speech-frame-threshold",
         type=int,
-        default=8,
+        default=5,
         help="Number of 20ms frames of speech before speech detection is triggered",
     )
     apps_run_parser.set_defaults(

{jaxl_python-0.0.37 → jaxl_python-0.0.38}/jaxl/api/resources/silence.py

@@ -21,7 +21,7 @@ class SilenceDetector:
         frame_duration_ms: int = 20,
         aggressiveness: int = 2,
         silence_frame_threshold: int = 12,  # ~240ms
-        speech_frame_threshold: int = 8,  # ~160ms
+        speech_frame_threshold: int = 5,  # ~100ms
     ):
         import webrtcvad
 

{jaxl_python-0.0.37 → jaxl_python-0.0.38}/jaxl_python.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: jaxl-python
-Version: 0.0.37
+Version: 0.0.38
 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-0.0.37 → jaxl_python-0.0.38}/jaxl_python.egg-info/SOURCES.txt

@@ -262,4 +262,5 @@ jaxl_python.egg-info/dependency_links.txt
 jaxl_python.egg-info/entry_points.txt
 jaxl_python.egg-info/not-zip-safe
 jaxl_python.egg-info/requires.txt
-jaxl_python.egg-info/top_level.txt
+jaxl_python.egg-info/top_level.txt
+tests/test_streaming_vad.py

jaxl_python-0.0.38/tests/test_streaming_vad.py

@@ -0,0 +1,113 @@
+"""
+Copyright (c) 2010-present by Jaxl Innovations Private Limited.
+
+All rights reserved.
+
+Redistribution and use in source and binary forms,
+with or without modification, is strictly prohibited.
+"""
+
+import base64
+import json
+import sys
+import types
+from typing import List
+
+import unittest
+
+from fastapi.testclient import TestClient
+
+from jaxl.api.base import BaseJaxlApp, JaxlStreamRequest
+from jaxl.api.resources.apps import _start_server
+from jaxl.api.resources.silence import SilenceDetector
+
+
+class _FakeVad:
+    def __init__(self, aggressiveness: int):
+        self._aggressiveness = aggressiveness
+
+    def is_speech(self, frame: bytes, sample_rate: int) -> bool:
+        return frame.startswith(b"\x01")
+
+
+class _StreamingCaptureApp(BaseJaxlApp):
+    def __init__(self) -> None:
+        self.speech_detection_events: List[bool] = []
+        self.speech_chunk_batches: List[List[bytes]] = []
+
+    async def handle_speech_detection(self, call_id: int, speaking: bool) -> None:
+        self.speech_detection_events.append(speaking)
+
+    async def handle_speech_chunks(
+        self,
+        req: JaxlStreamRequest,
+        slin16s: List[bytes],
+    ) -> None:
+        self.speech_chunk_batches.append(list(slin16s))
+
+
+def _install_fake_webrtcvad() -> None:
+    sys.modules["webrtcvad"] = types.SimpleNamespace(Vad=_FakeVad)
+
+
+def _media_event(payload: bytes) -> dict:
+    return {
+        "event": "media",
+        "media": {
+            "payload": base64.b64encode(payload).decode("ascii"),
+        },
+    }
+
+
+class StreamingVadTest(unittest.TestCase):
+    def setUp(self) -> None:
+        _install_fake_webrtcvad()
+
+    def test_silence_detector_uses_faster_speech_start_threshold(self) -> None:
+        detector = SilenceDetector()
+
+        changes = [detector.process(b"\x01" * 320) for _ in range(4)]
+        self.assertEqual(changes, [None, None, None, None])
+
+        change = detector.process(b"\x01" * 320)
+        self.assertTrue(change)
+
+    def test_streaming_vad_preserves_preroll_with_earlier_speech_start(self) -> None:
+        app = _StreamingCaptureApp()
+        server = _start_server(app, vad_speech_frame_threshold=5)
+        state = {
+            "call_id": 99,
+            "from_number": "+911111111111",
+            "to_number": "+922222222222",
+            "direction": 1,
+            "org": None,
+            "metadata": None,
+            "greeting_message": None,
+            "options": None,
+            "voice": None,
+            "campaign_id": None,
+            "campaign_type": None,
+        }
+        encoded_state = base64.urlsafe_b64encode(
+            json.dumps(state).encode("utf-8")
+        ).decode("ascii")
+
+        with TestClient(server) as client:
+            with client.websocket_connect(
+                f"/stream/?ivr_id=1&state={encoded_state}"
+            ) as websocket:
+                for _ in range(5):
+                    websocket.send_text(json.dumps(_media_event(b"\x01" * 320)))
+                for _ in range(12):
+                    websocket.send_text(json.dumps(_media_event(b"\x00" * 320)))
+
+        self.assertEqual(app.speech_detection_events[:2], [True, False])
+        self.assertGreaterEqual(len(app.speech_chunk_batches), 1)
+        self.assertEqual(len(app.speech_chunk_batches[0]), 4)
+        self.assertTrue(
+            all(chunk == b"\x01" * 320 for chunk in app.speech_chunk_batches[0])
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()