agora-python-server-sdk 2.3.4__tar.gz → 2.4.0__tar.gz

{agora_python_server_sdk-2.3.4/agora_python_server_sdk.egg-info → agora_python_server_sdk-2.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: agora_python_server_sdk
-Version: 2.3.4
+Version: 2.4.0
 Summary: A Python SDK for Agora Server
 Home-page: https://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK
 Classifier: Intended Audience :: Developers
@@ -64,6 +64,25 @@ python agora_rtc/examples/example_audio_pcm_send.py --appId=xxx --channelId=xxx
 ```
 
 # Change log
+## 2025.11.18 Release version 2.4.0
+-- Update SDK to APM version: 4.4.32/1025
+-- Add configure support for enabling/disabling APM
+-- Update methods in setup.py and __init__.py, ok, including version/URL, md5, etc.
+-- Overall code pipeline ok, needs testing?? ok
+-- Add support to rtm, one single sdk supports both rtm and rtc, ok
+-- todo:
+  -[] Need to add VAD algorithm update ok
+  -[] Need to add vad_dump modifications ok
+  -[] Modify APM algorithm to support VAD switch, ok
+  -[] Add VAD configure parameter settings, ok
+  -[] Download every time, check md5 mismatch?? ok
+  NOTE:
+  APM features, i.e., server-side echo cancellation, noise suppression, automatic gain control, background voice removal, etc.
+  Normally, AEC/AINS/AGC, etc., are already implemented on the client side, and the server side does not need to implement them again, unless there are special requirements.
+  If you want to enable APM features, please contact Agora technical support.
+  NOTE：
+  How to use rtc,please ref to: https://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK/tree/main/agora_rtc/examples
+  Hot to use rtm,please ref to ://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK/tree/main/agora_rtc/rtm_examples
 # 2025.11.07 release 2.3.3
 -- update: to support rtm
 -- adjust sdk's directory structure

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/README.md

@@ -49,6 +49,25 @@ python agora_rtc/examples/example_audio_pcm_send.py --appId=xxx --channelId=xxx
 ```
 
 # Change log
+## 2025.11.18 Release version 2.4.0
+-- Update SDK to APM version: 4.4.32/1025
+-- Add configure support for enabling/disabling APM
+-- Update methods in setup.py and __init__.py, ok, including version/URL, md5, etc.
+-- Overall code pipeline ok, needs testing?? ok
+-- Add support to rtm, one single sdk supports both rtm and rtc, ok
+-- todo:
+  -[] Need to add VAD algorithm update ok
+  -[] Need to add vad_dump modifications ok
+  -[] Modify APM algorithm to support VAD switch, ok
+  -[] Add VAD configure parameter settings, ok
+  -[] Download every time, check md5 mismatch?? ok
+  NOTE:
+  APM features, i.e., server-side echo cancellation, noise suppression, automatic gain control, background voice removal, etc.
+  Normally, AEC/AINS/AGC, etc., are already implemented on the client side, and the server side does not need to implement them again, unless there are special requirements.
+  If you want to enable APM features, please contact Agora technical support.
+  NOTE：
+  How to use rtc,please ref to: https://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK/tree/main/agora_rtc/examples
+  Hot to use rtm,please ref to ://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK/tree/main/agora_rtc/rtm_examples
 # 2025.11.07 release 2.3.3
 -- update: to support rtm
 -- adjust sdk's directory structure

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/__init__.py

@@ -122,11 +122,20 @@ def _check_download_and_extract_sdk():
     #fusion version: 20251023
 
     url = "https://download.agora.io/sdk/release/agora_rtc_sdk-x86_64-linux-gnu-v4.4.32-20250829_160340-860733-aed_20251107_1642.zip"
-
+    #20251110 Fusion version: with apm filter
+    mac_sdk="https://download.agora.io/sdk/release/agora_sdk_mac_v4.4.30_25869_FULL_20251030_1836_953684-aed.zip"
+    linux_sdk = "https://download.agora.io/sdk/release/agora_rtc_sdk_x86_64-linux-gnu-v4.4.32.150_26715_SERVER_20251030_1807-aed.zip"
+    
           
-    rtc_libfile_path = os.path.join(sdk_library_dir, "libagora_rtc_sdk.so")
+    linux_libfile_path = os.path.join(sdk_library_dir, "libagora_rtc_sdk.so")
+    mac_libfile_path = os.path.join(sdk_library_dir, "libAgoraRtcKit.dylib")
+    linux_md5 = "821cb1a388279648fcb204ca795e6476"
+    mac_md5 = "5b9940d3fca033a53ac30216d5c39be6"
+
     #rtc_md5 = "7031dd10d1681cd88fd89d68c5b54282"
-    rtc_md5 = "7eb8042e43246f95f188549d8711d1bf"
+    url = linux_sdk
+    rtc_md5 = linux_md5
+    rtc_libfile_path = linux_libfile_path
     if sys.platform == 'darwin':
         #url = "https://download.agora.io/sdk/release/agora_rtc_sdk_mac_rel.v4.4.30_22472_FULL_20241024_1224_398653.zip"
         # version   2.2.0 for mac
@@ -134,13 +143,12 @@ def _check_download_and_extract_sdk():
         #url = "https://download.agora.io/sdk/release/agora_sdk_mac_v4.4.32_24915_FULL_20250715_1710_791284.zip"
         #url = "https://download.agora.io/sdk/release/agora_sdk_mac_v4.4.32_25418_FULL_20250829_1647_860754.zip"
         #20251023 Fusion version: one sdk package include rtc and rtm
-        url = "https://download.agora.io/sdk/release/agora_sdk_mac_v4.4.32_25418_FULL_20250829_1647_860754-aed_20251107_1639.zip"
+        #url = "https://download.agora.io/sdk/release/agora_sdk_mac_v4.4.32_25418_FULL_20250829_1647_860754-aed_20251107_1639.zip"
+        url = mac_sdk
 
-        
-
-        rtc_libfile_path = os.path.join(sdk_library_dir, "libAgoraRtcKit.dylib")
+        rtc_libfile_path = mac_libfile_path
         #rtc_md5 = "ca3ca14f9e2b7d97eb2594d1f32dab9f"
-        rtc_md5 = "df0ec3b5073d17dee76cc4d97c13699a"
+        rtc_md5 = mac_md5
     if arch == "aarch64" and sys.platform == 'linux':
         #url = "https://download.agora.io/sdk/release/Agora-RTC-aarch64-linux-gnu-v4.4.31-20250307_175457-603878.zip"
         #url = "https://download.agora.io/sdk/release/Agora-RTC-aarch64-linux-gnu-v4.4.32-20250425_150503-675674.zip"
@@ -198,5 +206,6 @@ else:
 if not os.path.exists(rtc_libfile_path):
     logger.error(f"library {rtc_libfile_path} not found")
     sys.exit(1)
-# import these variables to the sub modules
-__all__ = ['sdk_library_dir', 'sdk_rtc_dir', 'sdk_rtm_dir', 'sdk_root_dir']
+
+# 显式导出这些变量，确保子模块可以导入
+__all__ = ['sdk_library_dir', 'sdk_rtc_dir', 'sdk_rtm_dir', 'sdk_root_dir']

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/__init__.py

@@ -22,11 +22,17 @@ try:
     if sys.platform == 'darwin':
         lib_agora_rtc_path = os.path.join(lib_dir, 'libAgoraRtcKit.dylib')
         agora_lib = ctypes.CDLL(lib_agora_rtc_path)
+        ctypes.CDLL(os.path.join(lib_dir, 'libAgoraAiNoiseSuppressionExtension.dylib'))
 
     elif sys.platform == 'linux':
         lib_agora_rtc_path = os.path.join(lib_dir, 'libagora_rtc_sdk.so')
         ctypes.CDLL(os.path.join(lib_dir, 'libagora-fdkaac.so'))
+        #ctypes.CDLL(os.path.join(lib_dir, 'libagora_ai_noise_suppression_extension.so'))
+        ctypes.CDLL(os.path.join(lib_dir, 'libagora-ffmpeg.so'))
+        ctypes.CDLL(os.path.join(lib_dir, 'libagora-soundtouch.so'))
         agora_lib = ctypes.CDLL(lib_agora_rtc_path)
+        # should load it or the ains can not work
+        ctypes.CDLL(os.path.join(lib_dir, 'libagora_ai_noise_suppression_extension.so'))
 except OSError as e:
     logger.error(f"Error loading the library: {e}")
     logger.error(f"Attempted to load from: {lib_agora_rtc_path}")

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/_ctypes_handle/_local_user_observer.py

@@ -186,6 +186,8 @@ class RTCLocalUserObserverInner(ctypes.Structure):
     def _on_user_audio_track_subscribed(self, local_user_handle, user_id, remote_audio_track_handle):
         logger.debug(f"LocalUserCB _on_user_audio_track_subscribed: {local_user_handle}, {user_id}, {remote_audio_track_handle}")
         user_id_str = user_id.decode('utf-8') if user_id else ""
+        #add apm filter for remote audio track
+        self.local_user._set_apm_filter_properties(remote_audio_track_handle,user_id_str)
         # note: this is a pointer to agora::rtc::IRemoteAudioTrack
         remote_audio_track = RemoteAudioTrack(remote_audio_track_handle, user_id_str)
         # map to localuser to save reference

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/agora_base.py

@@ -297,6 +297,63 @@ class VideoFrame():
     alpha_buffer: bytearray = None
     alpha_mode: int = 0
 
+@dataclass(kw_only=True)
+class AiNsConfig:
+	ns_enabled: bool = True
+	ai_ns_enabled: bool = True
+	ai_ns_model_pref: int = 10
+	nsng_alg_route: int = 12
+	nsng_predef_agg: int = 11
+@dataclass(kw_only=True)
+class AiAecConfig:
+	enabled: bool = False
+	split_srate_for_48k: int = 16000
+
+@dataclass(kw_only=True)
+class BghvsCConfig:
+	enabled: bool = True
+	vad_thr: float = 0.8
+
+@dataclass(kw_only=True)
+class AgcConfig:
+	enabled: bool = False
+
+@dataclass(kw_only=True)
+class APMConfig:
+    ai_ns_config: AiNsConfig = field(default_factory=AiNsConfig)
+    ai_aec_config: AiAecConfig = field(default_factory=AiAecConfig)
+    bghvs_c_config: BghvsCConfig = field(default_factory=BghvsCConfig)
+    agc_config: AgcConfig = field(default_factory=AgcConfig)
+    enable_dump: bool = False
+    
+    def _to_json_string(self):
+        import json
+        config_dict = {
+            "aec": {
+                "enabled": self.ai_aec_config.enabled,
+                "split_srate_for_48k": self.ai_aec_config.split_srate_for_48k
+            },
+            "bghvs": {
+                "enabled": self.bghvs_c_config.enabled,
+                "vadThr": self.bghvs_c_config.vad_thr
+            },
+            "agc": {
+                "enabled": self.agc_config.enabled
+            },
+            "ans": {
+                "enabled": self.ai_ns_config.ns_enabled
+            },
+            "sf_st_cfg": {
+                "enabled": self.ai_ns_config.ai_ns_enabled,
+                "ainsModelPref": self.ai_ns_config.ai_ns_model_pref
+            },
+            "sf_ext_cfg": {
+                "nsngAlgRoute": self.ai_ns_config.nsng_alg_route,
+                "nsngPredefAgg": self.ai_ns_config.nsng_predef_agg
+            }
+        }
+        return json.dumps(config_dict, separators=(',', ':'))
+    
 
 @dataclass(kw_only=True)
 class AgoraServiceConfig:
@@ -324,6 +381,11 @@ class AgoraServiceConfig:
     log_file_size_kb: int = 5*1024
     data_dir: str = ""
     config_dir: str = "" #format like: "./agora_rtc_log"
+    #20251110 Fusion version: with apm filter
+    enable_apm: bool = False
+    apm_config: APMConfig = None
+
+
 
 
 @dataclass(kw_only=True)
@@ -566,4 +628,4 @@ class CapabilityItemMap:
 @dataclass(kw_only=True)
 class Capabilities:
     item_map: CapabilityItemMap = None
-    capability_type: int = 0
+    capability_type: int = 0

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/agora_service.py

@@ -65,6 +65,14 @@ agora_service_set_log_filter = agora_lib.agora_service_set_log_filter
 agora_service_set_log_filter.restype = AGORA_API_C_INT
 agora_service_set_log_filter.argtypes = [AGORA_HANDLE, ctypes.c_uint]
 
+agora_audio_track_enable_audio_filter = agora_lib.agora_audio_track_enable_audio_filter
+agora_audio_track_enable_audio_filter.restype = ctypes.c_int
+agora_audio_track_enable_audio_filter.argtypes = [AGORA_HANDLE, ctypes.c_char_p, ctypes.c_int, ctypes.c_int]
+
+agora_audio_track_set_filter_property = agora_lib.agora_audio_track_set_filter_property
+agora_audio_track_set_filter_property.restype = ctypes.c_int
+agora_audio_track_set_filter_property.argtypes = [AGORA_HANDLE, ctypes.c_char_p, ctypes.c_char_p, ctypes.c_char_p, ctypes.c_int]
+
 
 class AgoraService:
     def __init__(self) -> None:
@@ -74,6 +82,8 @@ class AgoraService:
         self.inited = False
         #default to None, and never create it manually by developer from ver2.3.0
         self.media_node_factory = None
+        self.enable_apm = False
+        self.apm_config = None
 
     def initialize(self, config: AgoraServiceConfig):
         if self.inited == True:
@@ -101,7 +111,14 @@ class AgoraService:
 
         # force audio vad v2 to be enabled
         agora_parameter.set_parameters("{\"che.audio.label.enable\": true}")
-
+        # for apm filter: to enable apm filter
+        generator = "audio_processing_remote_playback"
+        cgenerator = generator.encode('utf-8')
+        ctrak = ctypes.c_char_p(None)
+        result = agora_service_enable_extension(self.service_handle, cprovider, cgenerator, ctrak, 1)
+        if result != 0:
+            logger.error(f"Failed to enable audio processing remote playback filter. Error code: {result}")
+       
         #versio 2.2.0 for callback when muted
         if config.should_callbck_when_muted > 0:
             agora_parameter.set_parameters("{\"rtc.audio.enable_user_silence_packet\": true}")
@@ -112,6 +129,17 @@ class AgoraService:
         '''
         agora_parameter.set_parameters("{\"che.video.min_enc_level\": 0}")
 
+        #keep & save apm config
+        self.enable_apm = config.enable_apm
+        if self.enable_apm:
+            if config.apm_config is None:
+                self.apm_config = APMConfig()
+            else:
+                self.apm_config = config.apm_config
+        else:
+            self.apm_config = None
+        
+
         return result
 
     def release(self):
@@ -210,3 +238,27 @@ class AgoraService:
         else:
             logger.error(f"Failed to set log file. Error code: {result}")
         return result
+    #apm related:
+#apm related api
+def _get_audio_filter_position(is_local_track: bool = False) -> int:
+    if is_local_track:
+        return 3
+    return 2
+
+def _enable_audio_filter_by_track(track: any, name: str, enable: bool, is_local_track: bool) -> int:
+    if track is None:
+        return -1000
+    c_name = ctypes.c_char_p(name.encode('utf-8'))
+    c_enable = ctypes.c_int(0)
+    if enable:
+        c_enable = ctypes.c_int(1)
+    position = _get_audio_filter_position(is_local_track)
+    return int(agora_audio_track_enable_audio_filter(track, c_name, c_enable, ctypes.c_int(position)))
+def _set_filter_property_by_track(track: any, name: str, key: str, value: str, is_local_track: bool) -> int:
+    if track is None:
+        return -1000
+    c_name = ctypes.c_char_p(name.encode('utf-8'))
+    c_key = ctypes.c_char_p(key.encode('utf-8'))
+    c_value = ctypes.c_char_p(value.encode('utf-8'))
+    position = _get_audio_filter_position(is_local_track)
+    return int(agora_audio_track_set_filter_property(track, c_name, c_key, c_value, ctypes.c_int(position)))

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/local_user.py

@@ -624,3 +624,7 @@ class LocalUser:
         size = len(data)
         ret = agora_local_user_send_aduio_meta_data(self.user_handle, c_data, ctypes.c_size_t(size))
         return ret
+    def _set_apm_filter_properties(self, remote_audio_track_handle, user_id_str)->int:
+        ret = self.connection._set_apm_filter_properties(remote_audio_track_handle, user_id_str)
+        print(f"**********LocalUser _set_apm_filter_properties: {ret}")
+        return ret

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/rtc_connection.py

@@ -2,7 +2,7 @@ import time
 import ctypes
 
 from .agora_base import *
-from .agora_service import AgoraService
+from .agora_service import AgoraService, _set_filter_property_by_track
 from .local_user import LocalUser
 from .rtc_connection_observer import IRTCConnectionObserver
 from ._ctypes_handle._audio_frame_observer import AudioFrameObserverInner
@@ -514,6 +514,32 @@ class RTCConnection:
         if self._capabilities_observer_obj:
             self._capabilities_observer_obj = None
         return 0
+
+    def _set_apm_filter_properties(self, remote_audio_track_handle, user_id_str)->int:
+        ret = -1000
+        is_enabled = self.rtc_engine.enable_apm
+        if not is_enabled:
+            return -1001
+        apm_config = self.rtc_engine.apm_config
+        if apm_config is None:
+            return -1002
+        apm_config_json = apm_config._to_json_string()
+        if apm_config_json is None:
+            return -1003
+        #Load AINS resource
+        ret = _set_filter_property_by_track(remote_audio_track_handle, "audio_processing_remote_playback", "apm_load_resource", "ains", False)
+        
+        print(f"**********APM: to set apm_load_resource, error: {ret}")
+        
+
+	    #Set APM configuration
+        ret = _set_filter_property_by_track(remote_audio_track_handle, "audio_processing_remote_playback", "apm_config", apm_config_json, False)
+        print(f"**********APM: to set apm_config, error: {ret}, apm_config_json: {apm_config_json}")
+        #Enable dump
+        if apm_config.enable_dump:
+            ret = _set_filter_property_by_track(remote_audio_track_handle, "audio_processing_remote_playback", "apm_dump", "true", False)
+            print(f"**********APM: to enable apm_dump, error: {ret}")
+        return ret
     
     
     

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/utils/vad_dump.py

@@ -3,6 +3,7 @@ import time
 from datetime import datetime
 import logging
 import os
+import struct
 from agora.rtc.agora_base import AudioFrame
 logger = logging.getLogger(__name__)
 
@@ -11,6 +12,42 @@ logger = logging.getLogger(__name__)
 """
 ## VadDump helper class
 """
+#buffer manager for vaddump: high performance buffer manager for vaddump
+
+class BufferManager:
+    """High-performance buffer manager using struct.pack
+    
+    Performance benchmarks (320 bytes, 100k iterations):
+    - struct.pack: 0.016-0.019 seconds (FASTEST)
+    - ctypes.memset: 0.107 seconds (5.6x slower)
+    - ctypes loop: 0.686 seconds (36x slower)
+    """
+    
+    def __init__(self, initial_size=320):
+        self.size = initial_size
+        self.buffer = bytearray(self.size)
+    def resize(self, new_size: int):
+        if new_size != self.size:
+            self.size = new_size
+            self.buffer = bytearray(self.size)
+    def fill_zero(self):
+        self.fill_int16(0)
+    
+    def fill_int16(self, value: int):
+        """Fill buffer with int16 value using struct.pack (fastest method)
+        
+        Args:
+            value: int16 value to fill (-32768 to 32767)
+        
+        Returns:
+            bytearray: The filled buffer
+        """
+        # Pack value as little-endian int16 and repeat
+        pattern = struct.pack('<h', value)
+        self.buffer[:] = pattern * (self.size // 2)
+        return self.buffer
+
+
 class VadDump():
     def __init__(self, path: str) -> None:
         self._file_path = path
@@ -20,6 +57,10 @@ class VadDump():
         self._source_file = None
         self._label_file = None
         self._vad_file = None
+        self._voice_prob_file = None
+        self._rms_file = None
+        self._pitch_file = None
+        self._buffer_manager = BufferManager(320)
         #check path is existed or not? if not, create new dir
         if self._check_directory_exists(path) is False:
             os.makedirs(path)
@@ -56,6 +97,16 @@ class VadDump():
         #open label file
         label_file_path = self._file_path + "/label.txt"
         self._label_file = open(label_file_path, "w")
+        #rms
+        rms_file_path = self._file_path + "/rms.pcm"
+        self._rms_file = open(rms_file_path, "wb")
+        #pitch
+        pitch_file_path = self._file_path + "/pitch.pcm"
+        self._pitch_file = open(pitch_file_path, "wb")
+        #voice prob
+        voice_prob_file_path = self._file_path + "/voice_prob.pcm"
+        self._voice_prob_file = open(voice_prob_file_path, "wb")
+
         #open vad file
         pass
     def write(self, frame:AudioFrame, vad_result_bytes: bytearray, vad_result_state : int) -> None:
@@ -68,6 +119,20 @@ class VadDump():
         if self._label_file:
             label_str = "ct:%d fct:%d state:%d far:%d vop:%d rms:%d pitch:%d mup:%d\n" % (self._count, self._frame_count,vad_result_state, frame.far_field_flag, frame.voice_prob, frame.rms, frame.pitch, frame.music_prob)
             self._label_file.write(label_str)
+        #adjut buffer size if needed
+        self._buffer_manager.resize(len(frame.buffer))
+        #write rms to buffer
+        self._buffer_manager.fill_int16(frame.rms*127)
+        if self._rms_file:
+            self._rms_file.write(self._buffer_manager.buffer)
+        #write pitch to buffe
+        self._buffer_manager.fill_int16(frame.pitch)
+        if self._pitch_file:
+            self._pitch_file.write(self._buffer_manager.buffer)
+        #write voice prob to buffer
+        self._buffer_manager.fill_int16(frame.voice_prob*127*127)
+        if self._voice_prob_file:
+            self._voice_prob_file.write(self._buffer_manager.buffer)
         #write to vad result
         if vad_result_state == 1: # start speaking
             #open new vad file and write header
@@ -96,6 +161,19 @@ class VadDump():
             self._label_file = None
         self._close_vad_file()
 
+        if self._source_file:
+            self._source_file.close()
+            self._source_file = None
+        if self._rms_file:
+            self._rms_file.close()
+            self._rms_file = None
+        if self._pitch_file:
+            self._pitch_file.close()
+            self._pitch_file = None
+        if self._voice_prob_file:
+            self._voice_prob_file.close()
+            self._voice_prob_file = None
+
         # assign to None
         self._count = 0
         self._frame_count = 0

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/agora/rtc/voice_detection.py

@@ -10,8 +10,8 @@ import logging
 logger = logging.getLogger(__name__)
 
 class  AudioVadConfigV2():
-    def __init__(self, preStartRecognizeCount:int, startRecognizeCount:int, stopRecognizeCount:int, 
-                 activePercent:float, inactivePercent:float, start_voiceprob: int, stop_voiceporb:int, rmsThreshold:float):
+    def __init__(self, preStartRecognizeCount:int=16, startRecognizeCount:int=30, stopRecognizeCount:int=50, 
+                 activePercent:float=0.7, inactivePercent:float=0.5, start_voiceprob: int=70, stop_voiceporb:int=50, rmsThreshold:int=-50):
         self.start_recognize_count = startRecognizeCount
         self.pre_start_recognize_count = preStartRecognizeCount
         self.stop_recognize_count = stopRecognizeCount
@@ -31,8 +31,12 @@ class  AudioVadConfigV2():
         # In a quiet environment, it can be set to -50;   
         # in a noisy environment, it can be set to a value between -40 and -30.
 
-        self.start_rms = rmsThreshold #default to -50
-        self.stop_rms = rmsThreshold #default to -50
+        self.start_rms = rmsThreshold #default to -70
+        self.stop_rms = rmsThreshold #default to -70
+        # for apm mode:
+        self.enable_adaptive_rms_threshold : bool = True
+        self.adaptive_rms_threshold_factor : float = 0.67 #default to 0.67, and 2/3 to pass
+        print(f"vadconfigure start_rms: {self.start_rms}, stop_rms: {self.stop_rms}")
         
         pass
 
@@ -63,11 +67,17 @@ class AudioVadV2():
         #trend queue: not impl in this version date: 2024-10-29
         self._trend_queue = None #deque(maxlen=self._vad_configure.stop_recognize_count)  
         self._trend_window = self._vad_configure.stop_recognize_count//2
+        self._voice_count : int = 0
+        self._silence_count : int = 0
+        self._total_voice_rms : int = 0
+        self._ref_avg_rms_in_last_session : int = 0
+
+        #convert rms from -120, 0 to range from 0 to 127, respond to db: -127db, to 0db
+        self._vad_configure.start_rms = 127 + self._vad_configure.start_rms
+        self._vad_configure.stop_rms = 127 + self._vad_configure.stop_rms
+        print(f"vad instance start_rms: {self._vad_configure.start_rms}, stop_rms: {self._vad_configure.stop_rms}")
+        pass
 
- 
-        
-        
-    
     def _push_to_start(self, data: VadDataV2) -> tuple[int,bool]:
         self._start_queue.append(data)
         size = len(self._start_queue)
@@ -140,19 +150,30 @@ class AudioVadV2():
         size, full = self._push_to_start(data)
         state = self._cur_state
         bytes = bytearray()
+
+        if data._is_activity == True:
+            self._voice_count += 1
+            self._total_voice_rms += data._audio_frame.rms
+        else:
+            self._voice_count = 0
+            self._total_voice_rms = 0
+        
+        
        
         
         if full == True:
-            #存在一定的问题：如果pre中就已经是开始在说话了，这个时候就会出现问题，或者漏掉的情况
-            #检查start中的比例是否符合阈值,如果符合阈值，zhi，则将start中的数据全部送入到pre中，并且将pre清空，同时将start清空，同时将当前状态设置为speaking
-            total, silence_count = self._get_silence_count(self._start_queue, self._vad_configure.pre_start_recognize_count)
-            total -= self._vad_configure.pre_start_recognize_count
-            if (total - silence_count) / total >= self._vad_configure.activePercent:
+            if self._voice_count >= self._vad_configure.start_recognize_count:
                 state = self._vad_state_startspeaking
                 #move pre & start to a new bytearray
                 
                 self._move_deque(bytes, self._start_queue)
                 self._clear_queue(self._start_queue)
+
+                #update ref
+                self._ref_avg_rms_in_last_session = int(self._total_voice_rms / self._voice_count)
+                self._voice_count = 0
+                self._total_voice_rms = 0
+                self._silence_count = 0
                
                 #and clear pre &start
                 self._clear_queue(self._stop_queue)
@@ -167,16 +188,19 @@ class AudioVadV2():
         size, full = self._push_to_stop(data)
         #print(f"stop: {size}, {full}")
 
+        if data._is_activity == True:
+            self._silence_count = 0
+        else:
+            self._silence_count += 1
+
         
-        if full == True:
-            #trend check
-            trend = self._get_trend(self._stop_queue)
-            #检查stop中的比例是否符合阈值,
-            #   如果符合阈值，同时清空stop 清空，并且将当前状态设置为non-speaking
-            total, silence_count = self._get_silence_count(self._stop_queue,0)
-            if (silence_count) / total >= (self._vad_configure.inactivePercent):
+        if full == True:    
+            if self._silence_count >= (self._vad_configure.stop_recognize_count):
                 state = self._vad_state_stopspeaking
                 self._clear_queue(self._stop_queue)
+                self._voice_count = 0
+                self._total_voice_rms = 0
+                self._silence_count = 0
                 #print(f"stop speaking: {len(self._start_queue)}, {silence_count}, {total}, {trend}")
         return state, data._audio_frame.buffer 
         
@@ -218,6 +242,11 @@ class AudioVadV2():
         #default: shoud never happen
         return int(-100), bytearray()
     
+    def _get_ref_active_avg_rms(self) -> int:
+        if self._vad_configure.enable_adaptive_rms_threshold and self._ref_avg_rms_in_last_session > 0:
+            return int(float(self._ref_avg_rms_in_last_session)*self._vad_configure.adaptive_rms_threshold_factor) #half of avg as threshold value
+        return self._vad_configure.start_rms
+       
     """
     def _is_vad_active(self, data: AudioFrame) -> bool:
     """
@@ -235,6 +264,11 @@ class AudioVadV2():
         #case2
         #if data.far_field_flag == 1 and data.voice_prob > voice_prob :#and data.pitch > 0 : #voice: from 75 to 50
         #case4: rms > -40
-        if data.far_field_flag == 1 and data.voice_prob > voice_prob and data.rms > rms_prob :#and data.pitch > 0 : #voice: from 75 to 50
+        #if data.far_field_flag == 1 and data.voice_prob > voice_prob and data.rms > rms_prob :#and data.pitch > 0 : #voice: from 75 to 50
+        #date: 2025-10-29 for sdk which support apm filter, and in this version, we don't need to use farfield flag to detect speech, so we don't need to use farfield flag to detect speech
+        refRms = self._get_ref_active_avg_rms()
+        active = (data.voice_prob == 1) or (data.rms > refRms)
+        #print(f"active: {active}, voice_prob: {data.voice_prob}, rms: {data.rms}, refRms: {refRms}")
+        if active:
             return True
         return False

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0/agora_python_server_sdk.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: agora_python_server_sdk
-Version: 2.3.4
+Version: 2.4.0
 Summary: A Python SDK for Agora Server
 Home-page: https://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK
 Classifier: Intended Audience :: Developers
@@ -64,6 +64,25 @@ python agora_rtc/examples/example_audio_pcm_send.py --appId=xxx --channelId=xxx
 ```
 
 # Change log
+## 2025.11.18 Release version 2.4.0
+-- Update SDK to APM version: 4.4.32/1025
+-- Add configure support for enabling/disabling APM
+-- Update methods in setup.py and __init__.py, ok, including version/URL, md5, etc.
+-- Overall code pipeline ok, needs testing?? ok
+-- Add support to rtm, one single sdk supports both rtm and rtc, ok
+-- todo:
+  -[] Need to add VAD algorithm update ok
+  -[] Need to add vad_dump modifications ok
+  -[] Modify APM algorithm to support VAD switch, ok
+  -[] Add VAD configure parameter settings, ok
+  -[] Download every time, check md5 mismatch?? ok
+  NOTE:
+  APM features, i.e., server-side echo cancellation, noise suppression, automatic gain control, background voice removal, etc.
+  Normally, AEC/AINS/AGC, etc., are already implemented on the client side, and the server side does not need to implement them again, unless there are special requirements.
+  If you want to enable APM features, please contact Agora technical support.
+  NOTE：
+  How to use rtc,please ref to: https://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK/tree/main/agora_rtc/examples
+  Hot to use rtm,please ref to ://github.com/AgoraIO-Extensions/Agora-Python-Server-SDK/tree/main/agora_rtc/rtm_examples
 # 2025.11.07 release 2.3.3
 -- update: to support rtm
 -- adjust sdk's directory structure

{agora_python_server_sdk-2.3.4 → agora_python_server_sdk-2.4.0}/setup.py

@@ -80,6 +80,12 @@ class CustomInstallCommand(install):
         url = "https://download.agora.io/sdk/release/agora_rtc_sdk-x86_64-linux-gnu-v4.4.32-20250829_160340-860733-aed_20251107_1642.zip"
         if sys.platform == 'darwin':
             url = "https://download.agora.io/sdk/release/agora_sdk_mac_v4.4.32_25418_FULL_20250829_1647_860754-aed_20251107_1639.zip"
+
+        #20251110 Fusion version: with apm filter
+        mac_sdk="https://download.agora.io/sdk/release/agora_sdk_mac_v4.4.30_25869_FULL_20251030_1836_953684-aed.zip"
+        url = "https://download.agora.io/sdk/release/agora_rtc_sdk_x86_64-linux-gnu-v4.4.32.150_26715_SERVER_20251030_1807-aed.zip"
+        if sys.platform == 'darwin':
+            url = mac_sdk
    
        
         if arch == "aarch64" and sys.platform == 'linux':
@@ -107,7 +113,7 @@ class CustomInstallCommand(install):
 
 setup(
     name='agora_python_server_sdk',
-    version='2.3.4',
+    version='2.4.0',
     description='A Python SDK for Agora Server',
     long_description=open('README.md').read(),
     long_description_content_type='text/markdown',