168 lines
6.3 KiB
Python
168 lines
6.3 KiB
Python
from fastapi import FastAPI, WebSocket, WebSocketDisconnect
|
|
import uvicorn
|
|
from starlette.middleware.cors import CORSMiddleware
|
|
import wave
|
|
import io
|
|
import asyncio
|
|
import websockets
|
|
import json
|
|
|
|
app = FastAPI()
|
|
|
|
# 跨域配置
|
|
app.add_middleware(
|
|
CORSMiddleware,
|
|
allow_origins=["*"],
|
|
allow_credentials=True,
|
|
allow_methods=["*"],
|
|
allow_headers=["*"],
|
|
)
|
|
|
|
# 存储活跃连接
|
|
active_connections = []
|
|
|
|
# 音频参数（必须和前端一致！）
|
|
AUDIO_PARAMS = {
|
|
"sample_rate": 16000, # 前端设置的 16000
|
|
"channels": 1, # 单声道
|
|
"sample_width": 2, # 16位深度（2字节/样本）
|
|
"frame_size": 1024 # 前端设置的 1024
|
|
}
|
|
|
|
# 语音识别服务配置（替换为你的识别服务地址）
|
|
ASR_CONFIG = {
|
|
# "host": "192.168.109.129",
|
|
"host": "10.10.10.202",
|
|
"port": 10096,
|
|
"mode": "2pass", # 识别模式：online/offline/2pass
|
|
"chunk_size": [5, 10, 5],
|
|
"chunk_interval": 10,
|
|
"use_itn": 1,
|
|
"hotwords": "" # 热词（可选）
|
|
}
|
|
|
|
# 封装 PCM 为 WAV（如果识别服务需要 WAV 格式）
|
|
def pcm_to_wav(pcm_data: bytes, params: dict) -> bytes:
|
|
output = io.BytesIO()
|
|
with wave.open(output, 'wb') as wf:
|
|
wf.setnchannels(params["channels"])
|
|
wf.setsampwidth(params["sample_width"])
|
|
wf.setframerate(params["sample_rate"])
|
|
wf.writeframes(pcm_data)
|
|
output.seek(0)
|
|
return output.read()
|
|
|
|
# 语音识别客户端协程：连接 ASR 服务，发送音频，接收结果
|
|
async def asr_client(websocket: WebSocket, audio_queue: asyncio.Queue):
|
|
"""
|
|
连接语音识别服务，处理音频转发和结果回传
|
|
:param websocket: 前端的 WebSocket 连接
|
|
:param audio_queue: 音频数据队列（接收前端的 PCM 数据）
|
|
"""
|
|
# 1. 连接 ASR 服务
|
|
asr_uri = f"ws://{ASR_CONFIG['host']}:{ASR_CONFIG['port']}"
|
|
try:
|
|
async with websockets.connect(
|
|
asr_uri,
|
|
subprotocols=["binary"],
|
|
ping_interval=None
|
|
) as asr_ws:
|
|
# 2. 发送 ASR 初始化配置
|
|
init_msg = json.dumps({
|
|
"mode": ASR_CONFIG["mode"],
|
|
"chunk_size": ASR_CONFIG["chunk_size"],
|
|
"chunk_interval": ASR_CONFIG["chunk_interval"],
|
|
"wav_name": "frontend_stream",
|
|
"is_speaking": True,
|
|
"hotwords": ASR_CONFIG["hotwords"],
|
|
"itn": bool(ASR_CONFIG["use_itn"]),
|
|
"audio_fs": AUDIO_PARAMS["sample_rate"]
|
|
})
|
|
await asr_ws.send(init_msg)
|
|
print("已发送 ASR 初始化配置", init_msg)
|
|
|
|
# 3. 协程1：从队列读取前端音频，发送给 ASR 服务
|
|
async def send_audio():
|
|
while True:
|
|
try:
|
|
# 从队列获取前端的 PCM 数据
|
|
pcm_data = await audio_queue.get()
|
|
# 如果识别服务需要 WAV 格式，取消下面注释
|
|
# wav_data = pcm_to_wav(pcm_data, AUDIO_PARAMS)
|
|
# await asr_ws.send(wav_data)
|
|
# 否则直接发送 PCM 裸数据
|
|
await asr_ws.send(pcm_data)
|
|
await asyncio.sleep(0.005) # 控制发送速率
|
|
except Exception as e:
|
|
print(f"发送音频到 ASR 失败：{e}")
|
|
break
|
|
|
|
# 4. 协程2：接收 ASR 识别结果，回传给前端
|
|
async def recv_asr_result():
|
|
while True:
|
|
try:
|
|
asr_result = await asr_ws.recv()
|
|
asr_result_json = json.loads(asr_result)
|
|
# 仅保留核心识别结果（按需调整）
|
|
result = {
|
|
"text": asr_result_json.get("text", ""),
|
|
"mode": asr_result_json.get("mode", ""),
|
|
"timestamp": asr_result_json.get("timestamp", ""),
|
|
"is_final": asr_result_json.get("is_final", False)
|
|
}
|
|
# 回传给当前前端客户端2
|
|
print('result', result)
|
|
ORIGINAL_STR = "Hello 世界！123@#"
|
|
binary_data = ORIGINAL_STR.encode("utf-8")
|
|
print(f"原始字符串：{ORIGINAL_STR}")
|
|
print(f"转二进制后：{binary_data}（长度：{len(binary_data)} 字节）")
|
|
|
|
# 2. 发送二进制数据
|
|
await websocket.send_bytes(binary_data)
|
|
except Exception as e:
|
|
print(f"接收 ASR 结果失败：{e}")
|
|
break
|
|
|
|
# 5. 并发运行两个协程
|
|
await asyncio.gather(send_audio(), recv_asr_result())
|
|
|
|
except Exception as e:
|
|
print(f"ASR 客户端连接失败：{e}")
|
|
# 给前端返回错误
|
|
await websocket.send_json({"error": str(e), "text": ""})
|
|
|
|
@app.websocket("/ws/audio")
|
|
async def websocket_audio(websocket: WebSocket):
|
|
await websocket.accept()
|
|
active_connections.append(websocket)
|
|
print(f"客户端连接成功，当前连接数：{len(active_connections)}")
|
|
|
|
# 创建音频队列：前端音频 -> 队列 -> ASR 客户端
|
|
audio_queue = asyncio.Queue(maxsize=100) # 队列大小防止积压
|
|
|
|
# 启动 ASR 客户端协程
|
|
asr_task = asyncio.create_task(asr_client(websocket, audio_queue))
|
|
|
|
try:
|
|
while True:
|
|
# 1. 接收前端的裸 PCM 音频帧
|
|
pcm_data = await websocket.receive_bytes()
|
|
# 2. 将音频放入队列，交给 ASR 客户端处理
|
|
if not audio_queue.full():
|
|
audio_queue.put_nowait(pcm_data)
|
|
|
|
except WebSocketDisconnect:
|
|
active_connections.remove(websocket)
|
|
print(f"客户端断开，当前连接数：{len(active_connections)}")
|
|
except Exception as e:
|
|
if websocket in active_connections:
|
|
active_connections.remove(websocket)
|
|
print(f"处理音频失败：{e}")
|
|
finally:
|
|
# 停止 ASR 客户端协程
|
|
asr_task.cancel()
|
|
await asr_task
|
|
|
|
if __name__ == "__main__":
|
|
# 启动 FastAPI 服务
|
|
uvicorn.run(app, host="0.0.0.0", port=8000) |