[Feature] 调整VAD工作流程，规范VAD产出数据规范为 models/audiobinary中的AudioBinary_Chunk；完整测试LogicTrager VAD online流程。

src/functor/__init__.py

@@ -0,0 +1,4 @@
+from .vad_functor import VAD
+from .model_loader import load_models
+
+__all__ = ["VAD", "load_models"]

src/functor/audiochunk.py

@@ -7,6 +7,7 @@
 import numpy as np
 import logging
 from typing import List, Optional, Union
+from src.models import AudioBinary_Config
 
 # 配置日志
 logging.basicConfig(
@@ -20,22 +21,19 @@ class AudioChunk:
     
     def __init__(self, 
                 max_duration_ms: int = 1000*60*60*10,
-                sample_rate: int = 16000,
+                audio_config : AudioBinary_Config = None,
-                sample_width: int = 2,
+                ):
-                channels: int = 1):
         """
         初始化音频数据块管理器
         
         参数:
             max_duration_ms: 音频池最大留存时间(ms)，默认10小时
-            sample_rate: 采样率，默认16KHz
+            audio_config: 音频配置信息
-            sample_width: 采样位宽，默认16bit
-            channels: 通道数，默认1
         """
         # 音频参数
-        self.sample_rate = sample_rate  # 采样率：16KHz
+        self.sample_rate = audio_config.sample_rate if audio_config is not None else 16000   # 采样率：16KHz
-        self.sample_width = sample_width     # 采样位宽：16bit
+        self.sample_width = audio_config.sample_width if audio_config is not None else 2     # 采样位宽：16bit
-        self.channels = channels         # 通道数：单声道
+        self.channels = audio_config.channels if audio_config is not None else 1         # 通道数：单声道
         
         # 数据存储
         self._max_duration_ms = max_duration_ms
@@ -84,6 +82,18 @@ class AudioChunk:
             logger.error(f"添加音频数据块时出错: {e}")
             return False
     
+    def get_chunk_binary(self, start: int = 0, end: Optional[int] = None) -> Optional[bytes]:
+        """
+        获取指定索引的音频数据块
+        """
+        print("[AudioChunk] get_chunk_binary", start, end)
+        if start >= len(self._chunk):
+            return None
+        if end is None or end > len(self._chunk):
+            end = len(self._chunk)
+        data = b''.join(self._chunk)
+        return data[start:end]
+
     def get_chunk(self, start_ms: int = 0, end_ms: Optional[int] = None) -> Optional[bytes]:
         """
         获取指定时间范围的音频数据

src/functor/model_loader.py

@@ -35,6 +35,7 @@ def load_models(args):
         device=args.device,
         disable_pbar=True,
         disable_log=True,
+        disable_update=True,
     )
     
     # 2. 加载在线ASR模型
@@ -47,6 +48,7 @@ def load_models(args):
         device=args.device,
         disable_pbar=True,
         disable_log=True,
+        disable_update=True,
     )
     
     # 3. 加载VAD模型
@@ -59,6 +61,7 @@ def load_models(args):
         device=args.device,
         disable_pbar=True,
         disable_log=True,
+        disable_update=True,
     )
     
     # 4. 加载标点符号模型（如果指定）
@@ -72,6 +75,7 @@ def load_models(args):
             device=args.device,
             disable_pbar=True,
             disable_log=True,
+            disable_update=True,
         )
     else:
         models["punc"] = None

src/functor/vad_functor.py

@@ -0,0 +1,60 @@
+from funasr import AutoModel
+from typing import List, Dict, Any
+from src.models import VADResponse
+from src.models import AudioBinary_Config
+from src.functor.audiochunk import AudioChunk
+from src.models import AudioBinary_Chunk
+from typing import Callable
+
+class VAD:
+
+    def __init__(self,
+        VAD_model = None,
+        audio_config : AudioBinary_Config = None,
+        callback: Callable = None,
+        ):
+        # vad model
+        self.VAD_model = VAD_model
+        if self.VAD_model is None:
+            self.VAD_model = AutoModel(model="fsmn-vad", model_revision="v2.0.4", disable_update=True)
+        # audio config
+        self.audio_config = audio_config
+        # vad result
+        self.vad_result = VADResponse(time_chunk_index_callback=callback)
+        # audio binary poll
+        self.audio_chunk = AudioChunk(
+            audio_config=self.audio_config
+        )
+        self.cache = {}
+    
+    def push_binary_data(self,
+        binary_data: bytes,
+        ):
+        # 压入二进制数据
+        self.audio_chunk.add_chunk(binary_data)
+        # 处理音频块
+        res = self.VAD_model.generate(input=binary_data, 
+            cache=self.cache, 
+            chunk_size=self.audio_config.chunk_size,
+            is_final=False)
+        # print("VAD generate", res)
+        if len(res[0]["value"]):
+            self.vad_result += VADResponse.from_raw(res)
+    
+    def set_callback(self,
+        callback: Callable,
+        ):
+        self.vad_result.time_chunk_index_callback = callback
+
+    def process_vad_result(self, callback: Callable = None):
+        # 处理VAD结果
+        callback = callback if callback is not None else self.vad_result.time_chunk_index_callback
+        self.vad_result.process_time_chunk(
+            lambda x : callback(
+                AudioBinary_Chunk(
+                    start_time=x["start_time"],
+                    end_time=x["end_time"],
+                    chunk=self.audio_chunk.get_chunk(x["start_time"], x["end_time"])
+                )
+            )
+        )

src/logic_trager.py

@@ -5,7 +5,7 @@
 """
 
 import logging
-from typing import Any, Dict, Type
+from typing import Any, Dict, Type, Callable
 
 # 配置日志
 logging.basicConfig(
@@ -88,57 +88,62 @@ class AutoAfterMeta(type):
 5->6 __after__push_result_queue 调用回调函数
 """
 
+from src.functor import VAD
+from src.models import AudioBinary_Config
+from src.models import AudioBinary_Chunk
+from typing import List
+
 class LogicTrager(metaclass=AutoAfterMeta):
     """逻辑触发器类"""
     
     def __init__(self,
         audio_chunk_max_size: int = 1024 * 1024 * 10,
-        sample_rate: int = 16000,
+        audio_config: AudioBinary_Config = None,
-        channels: int = 1,
+        result_callback: Callable = None,
-        on_result_callback: Callable = None,
+        models: Dict[str, Any] = None,
     ):
         """初始化"""
         # 存储音频块
-        self._audio_chunk = []
+        self._audio_chunk : List[AudioBinary_Chunk] = []
         # 存储二进制数据
         self._audio_chunk_binary = b''
         self._audio_chunk_max_size = audio_chunk_max_size
         # 音频参数
-        self._sample_rate = sample_rate
+        self._audio_config = audio_config if audio_config is not None else AudioBinary_Config()
-        self._channels = channels
         # 结果队列
         self._result_queue = []
-        # 回调函数
+        # 聚合结果回调函数
-        self._on_result_callback = on_result_callback
+        self._aggregate_result_callback = result_callback
+        # 组件
+        self._vad = VAD(VAD_model = models.get("vad"), audio_config = self._audio_config)
+        self._vad.set_callback(self.push_audio_chunk)
+
+
         logger.info("初始化LogicTrager")
     
     def push_binary_data(self, chunk: bytes) -> None:
         """
-        添加音频块
+        压入音频块至VAD模块
         
         参数:
             chunk: 音频数据块
         """
-        if self._audio_chunk is None:
+        # print("LogicTrager push_binary_data", len(chunk))
-            logger.error("AudioChunk未初始化")
+        self._vad.push_binary_data(chunk)
-            return
+        self.__after__push_binary_data()
-        
-        self._audio_chunk_binary += chunk
-        logger.debug(f"添加音频块，大小: {len(chunk)}字节")
     
     def __after__push_binary_data(self) -> None:
         """
         添加音频块后处理
-        VAD检测，将检测到的VAD压入音频块
         """
-        # VAD检测
+        # print("LogicTrager __after__push_binary_data")
-        pass
+        self._vad.process_vad_result()
-            # 压入音频块 push_audio_chunk
     
-    def push_audio_chunk(self, chunk: bytes) -> None:
+    def push_audio_chunk(self, chunk: AudioBinary_Chunk) -> None:
         """
-        压入音频块
+        音频处理
         """
+        print("LogicTrager push_audio_chunk [{}ms:{}ms] (len={})".format(chunk.start_time, chunk.end_time, len(chunk.chunk)))
         self._audio_chunk.append(chunk)
 
     def __after__push_audio_chunk(self) -> None:

src/models/__init__.py

@@ -0,0 +1,3 @@
+from .audiobinary import AudioBinary_Config, AudioBinary_Chunk
+from .vad import VADResponse
+__all__ = ["AudioBinary_Config", "AudioBinary_Chunk", "VADResponse"]

src/models/audiobinary.py

@@ -0,0 +1,16 @@
+from pydantic import BaseModel, Field
+
+class AudioBinary_Config(BaseModel):
+    """二进制音频块配置信息"""
+    audio_data: bytes = Field(description="音频数据", default=None)
+    chunk_size: int = Field(description="块大小", default=100)
+    chunk_stride: int = Field(description="块步长", default=1600)
+    sample_rate: int = Field(description="采样率", default=16000)
+    sample_width: int = Field(description="采样位宽", default=2)
+    channels: int = Field(description="通道数", default=1)
+
+class AudioBinary_Chunk(BaseModel):
+    """音频块"""
+    start_time: int = Field(description="开始时间", default=0)
+    end_time: int = Field(description="结束时间", default=0)
+    chunk: bytes = Field(description="音频块", default=None)

src/models/vad.py

@@ -38,11 +38,16 @@ class VADResponse(BaseModel):
         """处理时间块"""
         # print("Enter process_time_chunk", self.time_chunk_index, len(self.time_chunk))
         while self.time_chunk_index < len(self.time_chunk) - 1:
-            if self.time_chunk[self.time_chunk_index].end != -1:
+            index = self.time_chunk_index
+            if self.time_chunk[index].end != -1:
+                x = {
+                    "start_time": self.time_chunk[index].start, 
+                    "end_time": self.time_chunk[index].end
+                }
                 if callback is not None:
-                    callback(self.time_chunk_index)
+                    callback(x)
                 elif self.time_chunk_index_callback is not None:
-                    self.time_chunk_index_callback(self.time_chunk_index)
+                    self.time_chunk_index_callback(x)
                 else:
                     print("[Warning] No callback available")
                 self.time_chunk_index += 1
@@ -125,3 +130,14 @@ class VADResponse(BaseModel):
                 result_str += f"[{value_item.start}:{value_item.end}]\n"
         return result_str
 
+    def __iter__(self):
+        return iter(self.time_chunk)
+    
+    def __next__(self):
+        return next(self.time_chunk)
+    
+    def __len__(self):
+        return len(self.time_chunk)
+    
+    def __getitem__(self, index):
+        return self.time_chunk[index]

test_main.py

@@ -1,4 +1,4 @@
-from tests.modelsuse import vad_model_use_online
+from tests.modelsuse import vad_model_use_online_logic
 
-vad_result = vad_model_use_online("tests/vad_example.wav")
+vad_result = vad_model_use_online_logic("tests/vad_example.wav")
 print(vad_result)

tests/modelsuse.py

@@ -1,6 +1,6 @@
 from funasr import AutoModel
 from typing import List, Dict, Any
-from src.pydantic_models import VADResponse
+from src.models import VADResponse
 import time
 
 def vad_model_use_online(file_path: str) -> List[Dict[str, Any]]:
@@ -32,6 +32,36 @@ def vad_model_use_online(file_path: str) -> List[Dict[str, Any]]:
     #     print(item)
     return vad_result
 
+def vad_model_use_online_logic(file_path: str) -> List[Dict[str, Any]]:
+    from src.logic_trager import LogicTrager
+    import soundfile
+
+    from src.config import parse_args
+    args = parse_args()
+
+    from src.functor.model_loader import load_models
+    models = load_models(args)
+
+    chunk_size = 200 # ms
+    from src.models import AudioBinary_Config
+    import soundfile
+
+    speech, sample_rate = soundfile.read(file_path)
+    chunk_stride = int(chunk_size * sample_rate / 1000)
+    audio_config = AudioBinary_Config(sample_rate=sample_rate, sample_width=2, channels=1, chunk_size=chunk_size)
+
+    logic_trager = LogicTrager(models=models, audio_config=audio_config)
+    for i in range(len(speech)//chunk_stride+1):
+        speech_chunk = speech[i*chunk_stride:(i+1)*chunk_stride]
+        logic_trager.push_binary_data(speech_chunk)
+
+    # for item in items:
+    #     print(item)
+    return None
+    
+    
+
 if __name__ == "__main__":
-    vad_result = vad_model_use_online("tests/vad_example.wav")
+    # vad_result = vad_model_use_online("tests/vad_example.wav")
+    vad_result = vad_model_use_online_logic("tests/vad_example.wav")
     # print(vad_result)