環境：

AI-Sphere-Butler

VBCABLE2.1.58

Win10專業版

豆包桌面版1.47.4

ubuntu22.04

英偉達4070ti 12G

python3.10

問題描述：

AI-Sphere-Butler之如何將豆包桌面版對接到AI全能管家~新玩法（一）

聊天視頻：

解決方案：

1.先安裝VBCABLE2.1.58工具，采集豆包音頻

“VBCABLE_Driver_Pack45win10” 指的是適用于 Windows 10 系統的 VBCABLE 驅動程序包，版本號可能是 45 。“VBCABLE” 可能是該驅動相關的產品或技術名稱，“Driver_Pack” 明確是驅動程序包，而 “win10” 表明其適用的操作系統為 Windows 10 。例如，可能是一種虛擬音頻電纜相關的驅動包，用于在 Windows 10 系統上實現特定音頻功能。

下載完軟件安裝x64版

繼續安裝

2.打開電腦聲音設置找到應用音量和設備首選項

3.將豆包程序的輸出設備選擇CABLEInput

4.自行安裝python和安裝依賴：

pip install flask flask-sockets gevent gevent-websocket

5.編寫采集豆包聲音客戶端

Collection.py文件內容：

import asyncio
import sounddevice as sd
import websockets
import numpy as np
import signal
import threading
import time
from collections import dequeINPUT_RATE = 16000
CHANNELS = 1
FRAME_SIZE = 640  
WS_URL = "ws://192.168.1.4:8020"#websockets服務地址
SILENCE_THRESHOLD = 1000stop_event = threading.Event()
signal.signal(signal.SIGINT, lambda s, f: stop_event.set())class AudioBuffer:def __init__(self, max_frames=20):self.buffer = deque(maxlen=max_frames)self.lock = threading.Lock()def put(self, frame_bytes):with self.lock:if len(self.buffer) == self.buffer.maxlen:self.buffer.popleft()print("[BUF] Buffer full, dropping oldest frame")self.buffer.append(frame_bytes)def get_all(self):with self.lock:frames = list(self.buffer)self.buffer.clear()return framesdef size(self):with self.lock:return len(self.buffer)def is_voice(data_np):energy = np.mean(data_np.astype(np.float32) ** 2)return energy > SILENCE_THRESHOLDdef audio_callback(indata, frames, time_info, status, audio_buffer):if status:print(f"[CAP] Warning: {status}")audio_np = indata[:, 0]ts = time.time()if is_voice(audio_np):frame = audio_np.tobytes()#print(f"[CAP] Voice frame captured at {ts:.3f}s, energy sufficient")else:frame = (np.zeros_like(audio_np)).tobytes()#print(f"[CAP] Silence frame at {ts:.3f}s")audio_buffer.put(frame)async def sender(ws, audio_buffer):while not stop_event.is_set():frames = audio_buffer.get_all()if not frames:await asyncio.sleep(0.005)continuefor frame in frames:try:await ws.send(frame)#print(f"[SND] Sent frame size={len(frame)} at {time.time():.3f}s, buffer size={audio_buffer.size()}")except Exception as e:print(f"[SND] Send error: {e}")stop_event.set()returnasync def capture_and_send(ws):audio_buffer = AudioBuffer(20)device_index = Nonedevices = sd.query_devices()for i, d in enumerate(devices):if "CABLE" in d['name'] and d['max_input_channels'] >= CHANNELS:device_index = ibreakif device_index is None:device_index = sd.default.device[0]print(f"[SYS] Using device #{device_index}: {devices[device_index]['name']}")send_task = asyncio.create_task(sender(ws, audio_buffer))with sd.InputStream(samplerate=INPUT_RATE,device=device_index,channels=CHANNELS,dtype='int16',blocksize=FRAME_SIZE,callback=lambda indata, frames, time_info, status:audio_callback(indata, frames, time_info, status, audio_buffer)):print("[SYS] Recording started.")while not stop_event.is_set():await asyncio.sleep(0.1)send_task.cancel()try:await send_taskexcept asyncio.CancelledError:passprint("[SYS] Recording stopped.")async def main():print(f"[SYS] Connecting to {WS_URL}")try:async with websockets.connect(WS_URL) as ws:print("[SYS] Connected.")await capture_and_send(ws)except Exception as e:print(f"[ERR] Connection error: {e}")if __name__ == '__main__':asyncio.run(main())

6.主程序引入模塊文件websocket_service.py：

AI-Sphere-Butler\core\server\virtual_human\websocket_service.py

import asyncio
import uuid
import websockets
import multiprocessing
import queueMAX_QUEUE_SIZE = 10def enqueue_audio_data(audio_queue, data):try:audio_queue.put_nowait(data)except queue.Full:try:discarded = audio_queue.get_nowait()print("[WSrv] 丟棄過舊音頻包，防止積壓")except queue.Empty:passtry:audio_queue.put_nowait(data)except queue.Full:# print("[WSrv] 隊列滿，丟棄當前音頻包")passasync def audio_handler(websocket, audio_queue: multiprocessing.Queue):session_id = str(uuid.uuid4())# print(f"[WSrv] Session {session_id} connected")try:async for raw in websocket:if isinstance(raw, (bytes, bytearray)):enqueue_audio_data(audio_queue, (session_id, raw))# print(f"[WSrv] Queued {len(raw)} bytes from {session_id}")else:# print(f"[WSrv] Ignored non-binary message from {session_id}")passexcept websockets.exceptions.ConnectionClosed:passfinally:# print(f"[WSrv] Session {session_id} disconnected")passasync def run_server(audio_queue: multiprocessing.Queue, host='0.0.0.0', port=8020):async def handler(websocket):await audio_handler(websocket, audio_queue)server = await websockets.serve(handler, host, port)# print(f"[WSrv] Listening on ws://{host}:{port}")await asyncio.Future()  if __name__ == "__main__":q = multiprocessing.Queue(maxsize=MAX_QUEUE_SIZE)asyncio.run(run_server(q))

7.運行采集客戶端和AI-Sphere-Butler服務

8.這樣就可以和豆包聊天，驅動AI全能管家數字人說話了