add banbanmini backend

talkingq-url/handlers/audio_packet_parser.py

@@ -0,0 +1,87 @@
+import struct
+from fastapi import WebSocket
+from services.audio_session import audio_session_manager, AudioSession
+from utils.logger import session_logger
+from services.factory import get_asr, get_llm, get_tts
+
+DEVICE_ID_SIZE = 22  # 将大小设为22，在实际解析中会更灵活处理
+SESSION_ID_SIZE = 33
+SAMPLE_RATE = 16000  # 固定为16K采样率
+CHANNELS = 1  # 单声道
+SAMPLE_WIDTH = 2  # 16位 = 2字节
+
+async def parse_packet(data: bytes, websocket: WebSocket):
+    try:
+        min_header_size = 4 + 1 + 4  # sequence_number + packet_type + data_size
+        if len(data) < min_header_size:
+            session_logger.warning(
+                "unknown", "unknown", f"数据包太小: {len(data)} < {min_header_size}"
+            )
+            return None, None, None, None, None
+            
+        null_pos = data.find(b'\x00')
+        if null_pos == -1 or null_pos > 25:  # 如果没找到或超出最大长度限制
+            null_pos = 22  # 使用默认值
+            
+        device_id = data[:null_pos].decode("ascii")
+        
+        session_start = null_pos + 1
+        session_end = session_start + SESSION_ID_SIZE
+        session_data = data[session_start:session_end]
+        session_null_pos = session_data.find(b'\x00')
+        
+        if session_null_pos != -1:
+            session_id = session_data[:session_null_pos].decode("ascii")
+        else:
+            session_id = session_data.decode("ascii").rstrip("\x00")
+            
+
+        offset = session_start + len(session_id) + 1
+        if offset + min_header_size > len(data):
+            session_logger.warning(
+                "unknown", "unknown", f"解析剩余头信息时数据包太小: {len(data)} < {offset + min_header_size}"
+            )
+            return None, None, None, None, None
+            
+        remaining_data = data[offset:]
+        offset = 0
+        sequence_number = struct.unpack("<I", remaining_data[offset:offset + 4])[0]
+        offset += 4
+        packet_type = remaining_data[offset]
+        offset += 1
+        data_size = struct.unpack("<I", remaining_data[offset:offset + 4])[0]
+        offset += 4
+        audio_data = remaining_data[offset:offset + data_size]
+
+        session_key = (device_id, session_id)
+        sample_rate = SAMPLE_RATE
+        session = await audio_session_manager.get_session(session_key)
+
+        if not session:
+            asr_service = await get_asr(None)
+            llm_service = await get_llm(None)
+            tts_service = await get_tts(None)
+            
+            session = AudioSession(
+                asr_service=asr_service,
+                llm_service=llm_service,
+                tts_service=tts_service,
+                sample_rate=sample_rate,
+            )
+            session.device_id = device_id
+            session.session_id = session_id
+            await audio_session_manager.set_session(session_key, session)
+
+        if packet_type == 1:
+            session.sample_rate = sample_rate
+
+        session.last_received_seq = max(session.last_received_seq, sequence_number)
+        return session_key, session, packet_type, audio_data, sample_rate
+    except Exception as e:
+        session_logger.error(
+            "unknown",
+            "unknown",
+            f"解析WebSocket数据包时出错: {str(e)}",
+            exc_info=True,
+        )
+        return None, None, None, None, None