John Qiu
712063071c
skills/ → skills-dev(9), skills-req(10), skills-ops(4), skills-integration(8), skills-biz(4), skills-workflow(7) generate-marketplace.py 改为自动扫描所有 skills-* 目录。 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
328 lines
12 KiB
Python
Executable File
328 lines
12 KiB
Python
Executable File
#!/usr/bin/env python3
|
||
"""
|
||
豆包唱歌工具
|
||
基于豆包端到端实时语音大模型，支持让豆包唱歌
|
||
使用WebSocket实时对话和生成音频
|
||
"""
|
||
|
||
import os
|
||
import sys
|
||
import json
|
||
import asyncio
|
||
import websockets
|
||
import struct
|
||
import uuid
|
||
from typing import Optional
|
||
|
||
|
||
# 连接级事件（不需要session_id）
|
||
CONNECTION_EVENTS = {1, 2, 50, 51, 52}
|
||
|
||
|
||
class DoubaoSinging:
|
||
"""豆包唱歌工具类"""
|
||
|
||
def __init__(self):
|
||
# 从环境变量读取配置
|
||
self.app_id = os.environ.get("DOUBAO_APP_ID")
|
||
self.access_token = os.environ.get("DOUBAO_ACCESS_TOKEN")
|
||
|
||
if not self.app_id or not self.access_token:
|
||
raise ValueError(
|
||
"请先设置环境变量:\n"
|
||
"export DOUBAO_APP_ID='your_app_id'\n"
|
||
"export DOUBAO_ACCESS_TOKEN='your_access_token'"
|
||
)
|
||
|
||
# 端到端实时语音WebSocket地址
|
||
self.ws_url = "wss://openspeech.bytedance.com/api/v3/realtime/dialogue"
|
||
self.app_key = "PlgvMymc7f3tQnJ6" # 固定值
|
||
self.resource_id = "volc.speech.dialog" # 固定值
|
||
|
||
def _build_message(self, event_id: int, payload: dict = None, session_id: str = None) -> bytes:
|
||
"""
|
||
构建二进制消息
|
||
|
||
协议格式:
|
||
- header (4 bytes)
|
||
- event_id (4 bytes, big-endian)
|
||
- [session_id_len (4 bytes) + session_id (variable)] -- 仅非连接级事件
|
||
- payload_len (4 bytes, big-endian)
|
||
- payload (variable, JSON)
|
||
"""
|
||
buf = bytearray()
|
||
|
||
# Header (4 bytes)
|
||
buf.append(0x11) # version=1, header_size=1
|
||
buf.append(0x14) # FULL_CLIENT_REQUEST(0x1) + WITH_EVENT(0x4)
|
||
buf.append(0x10) # JSON serialization, no compression
|
||
buf.append(0x00) # reserved
|
||
|
||
# Event ID
|
||
buf.extend(struct.pack('>I', event_id))
|
||
|
||
# Session ID (required for non-connection events)
|
||
if event_id not in CONNECTION_EVENTS:
|
||
sid_bytes = (session_id or "").encode('utf-8')
|
||
buf.extend(struct.pack('>I', len(sid_bytes)))
|
||
buf.extend(sid_bytes)
|
||
|
||
# Payload
|
||
if payload:
|
||
payload_bytes = json.dumps(payload, ensure_ascii=False).encode('utf-8')
|
||
else:
|
||
payload_bytes = b'{}'
|
||
buf.extend(struct.pack('>I', len(payload_bytes)))
|
||
buf.extend(payload_bytes)
|
||
|
||
return bytes(buf)
|
||
|
||
def _parse_response(self, data: bytes) -> dict:
|
||
"""
|
||
解析服务端二进制消息
|
||
|
||
Returns:
|
||
dict with keys: msg_type, event_id, session_id, payload, payload_bytes
|
||
"""
|
||
result = {"raw": data}
|
||
if len(data) < 4:
|
||
return result
|
||
|
||
# Header
|
||
msg_type = (data[1] >> 4) & 0x0F
|
||
flags = data[1] & 0x0F
|
||
result["msg_type"] = msg_type
|
||
|
||
offset = 4
|
||
|
||
# Event ID (if WITH_EVENT flag)
|
||
if flags & 0x04 and len(data) >= offset + 4:
|
||
event_id = struct.unpack('>I', data[offset:offset + 4])[0]
|
||
result["event_id"] = event_id
|
||
offset += 4
|
||
|
||
# Connect ID for connection events (50, 51, 52)
|
||
if event_id in {50, 51, 52} and len(data) >= offset + 4:
|
||
cid_len = struct.unpack('>I', data[offset:offset + 4])[0]
|
||
offset += 4
|
||
if len(data) >= offset + cid_len:
|
||
result["connect_id"] = data[offset:offset + cid_len].decode('utf-8', errors='ignore')
|
||
offset += cid_len
|
||
# Session ID for session-level events
|
||
elif event_id not in CONNECTION_EVENTS and len(data) >= offset + 4:
|
||
sid_len = struct.unpack('>I', data[offset:offset + 4])[0]
|
||
offset += 4
|
||
if len(data) >= offset + sid_len:
|
||
result["session_id"] = data[offset:offset + sid_len].decode('utf-8', errors='ignore')
|
||
offset += sid_len
|
||
|
||
# Payload
|
||
if len(data) >= offset + 4:
|
||
payload_len = struct.unpack('>I', data[offset:offset + 4])[0]
|
||
offset += 4
|
||
if len(data) >= offset + payload_len:
|
||
payload_raw = data[offset:offset + payload_len]
|
||
result["payload_bytes"] = payload_raw
|
||
# Audio-only responses (msg_type 0xB) have raw audio
|
||
if msg_type == 0x0B:
|
||
result["is_audio"] = True
|
||
else:
|
||
try:
|
||
result["payload"] = json.loads(payload_raw.decode('utf-8'))
|
||
except:
|
||
result["payload_text"] = payload_raw.decode('utf-8', errors='ignore')
|
||
|
||
return result
|
||
|
||
async def sing(
|
||
self,
|
||
song_request: str,
|
||
output_file: str = "singing_output.mp3",
|
||
language: str = "zh-CN",
|
||
model: str = "1.2.1.0"
|
||
) -> str:
|
||
"""
|
||
让豆包唱歌
|
||
|
||
Args:
|
||
song_request: 唱歌请求，如 "请唱一首关于春天的歌"
|
||
output_file: 输出音频文件路径
|
||
language: 语言代码 (zh-CN/en-US)
|
||
model: 模型版本
|
||
|
||
Returns:
|
||
str: 输出文件路径
|
||
"""
|
||
print(f"🎵 豆包唱歌中...")
|
||
print(f" 请求: {song_request}")
|
||
print(f" 模型: {model}")
|
||
|
||
try:
|
||
audio_data = bytearray()
|
||
session_id = str(uuid.uuid4())
|
||
|
||
# WebSocket连接头
|
||
headers = {
|
||
"X-Api-App-ID": self.app_id,
|
||
"X-Api-Access-Key": self.access_token,
|
||
"X-Api-Resource-Id": self.resource_id,
|
||
"X-Api-App-Key": self.app_key,
|
||
"X-Api-Connect-Id": str(uuid.uuid4()),
|
||
}
|
||
|
||
async with websockets.connect(self.ws_url, additional_headers=headers) as websocket:
|
||
print("✅ WebSocket连接成功")
|
||
|
||
# 1. StartConnection (event_id=1, 无需session_id)
|
||
await websocket.send(self._build_message(1))
|
||
response = await asyncio.wait_for(websocket.recv(), timeout=5)
|
||
resp = self._parse_response(response)
|
||
if resp.get("event_id") == 50:
|
||
print(f"✅ 连接已建立")
|
||
else:
|
||
print(f"⚠️ 连接响应: {resp}")
|
||
|
||
# 2. StartSession (event_id=100, 需要session_id)
|
||
start_session_payload = {
|
||
"tts": {
|
||
"audio_config": {
|
||
"channel": 1,
|
||
"format": "pcm",
|
||
"sample_rate": 24000
|
||
}
|
||
},
|
||
"dialog": {
|
||
"extra": {
|
||
"enable_music": True,
|
||
"input_mod": "text",
|
||
"model": model
|
||
}
|
||
}
|
||
}
|
||
await websocket.send(self._build_message(100, start_session_payload, session_id))
|
||
response = await asyncio.wait_for(websocket.recv(), timeout=5)
|
||
resp = self._parse_response(response)
|
||
if resp.get("event_id") == 150:
|
||
print(f"✅ 会话已建立")
|
||
elif resp.get("payload", {}).get("error"):
|
||
print(f"❌ 会话错误: {resp['payload']['error']}")
|
||
return None
|
||
else:
|
||
print(f"📋 会话响应: {resp}")
|
||
|
||
# 3. SayHello/ChatTextQuery (event_id=300, 需要session_id)
|
||
chat_payload = {"content": song_request}
|
||
await websocket.send(self._build_message(300, chat_payload, session_id))
|
||
print(f"📤 已发送唱歌请求")
|
||
|
||
# 4. 接收音频流（使用超时检测结束）
|
||
print("\n📋 接收音频流...")
|
||
tts_started = False
|
||
recv_timeout = 5 # 5秒无数据则认为结束
|
||
|
||
while True:
|
||
try:
|
||
message = await asyncio.wait_for(websocket.recv(), timeout=recv_timeout)
|
||
except asyncio.TimeoutError:
|
||
break
|
||
except websockets.exceptions.ConnectionClosed:
|
||
break
|
||
|
||
if isinstance(message, bytes) and len(message) >= 4:
|
||
resp = self._parse_response(message)
|
||
msg_type = resp.get("msg_type", 0)
|
||
flags = message[1] & 0x0F
|
||
|
||
# Audio-only response (0xB = 11)
|
||
if resp.get("is_audio") and resp.get("payload_bytes"):
|
||
audio_data.extend(resp["payload_bytes"])
|
||
if not tts_started:
|
||
print(f" 接收音频中...", end="", flush=True)
|
||
tts_started = True
|
||
else:
|
||
print(".", end="", flush=True)
|
||
|
||
# NEG_SEQUENCE flag = last packet
|
||
if flags & 0x02:
|
||
break
|
||
|
||
# Server error (0xF = 15)
|
||
elif msg_type == 0x0F:
|
||
error = resp.get("payload", {}).get("error", "unknown")
|
||
print(f"\n❌ 服务器错误: {error}")
|
||
break
|
||
|
||
# Full server response (0x9) - session finished
|
||
elif msg_type == 0x09:
|
||
event_id = resp.get("event_id", 0)
|
||
if event_id in {152, 52}:
|
||
break
|
||
|
||
# 5. 保存音频文件
|
||
if audio_data:
|
||
# Save as PCM, convert extension if needed
|
||
actual_output = output_file
|
||
if output_file.endswith('.mp3'):
|
||
actual_output = output_file.replace('.mp3', '.pcm')
|
||
|
||
with open(actual_output, "wb") as f:
|
||
f.write(audio_data)
|
||
|
||
file_size = len(audio_data) / 1024
|
||
print(f"\n\n✅ 唱歌完成!")
|
||
print(f" 输出: {actual_output} ({file_size:.1f} KB)")
|
||
print(f" 格式: PCM (24000Hz, 单声道)")
|
||
return actual_output
|
||
else:
|
||
print("\n⚠️ 未收到音频数据，请检查:")
|
||
print(" 1. 凭证是否正确")
|
||
print(" 2. 端到端实时语音大模型是否已开通")
|
||
print(" 3. 网络连接是否正常")
|
||
return None
|
||
|
||
except websockets.exceptions.WebSocketException as e:
|
||
raise Exception(f"WebSocket连接错误: {str(e)}")
|
||
except Exception as e:
|
||
raise Exception(f"唱歌调用失败: {str(e)}")
|
||
|
||
|
||
def main():
|
||
"""命令行工具"""
|
||
import argparse
|
||
|
||
parser = argparse.ArgumentParser(description="豆包唱歌工具")
|
||
subparsers = parser.add_subparsers(dest="command", help="选择功能")
|
||
|
||
# 唱歌命令
|
||
sing_parser = subparsers.add_parser("sing", help="让豆包唱歌")
|
||
sing_parser.add_argument("request", help="唱歌请求，如 '请唱一首关于春天的歌'")
|
||
sing_parser.add_argument(
|
||
"-o", "--output", default="singing_output.mp3", help="输出音频文件（默认: singing_output.mp3）"
|
||
)
|
||
sing_parser.add_argument(
|
||
"-l", "--language", default="zh-CN", help="语言代码（默认: zh-CN）"
|
||
)
|
||
sing_parser.add_argument(
|
||
"-m", "--model", default="1.2.1.0", help="模型版本（默认: 1.2.1.0=O2.0版本）"
|
||
)
|
||
|
||
args = parser.parse_args()
|
||
|
||
if not args.command:
|
||
parser.print_help()
|
||
return
|
||
|
||
try:
|
||
singing = DoubaoSinging()
|
||
|
||
if args.command == "sing":
|
||
asyncio.run(singing.sing(args.request, args.output, args.language, args.model))
|
||
|
||
except Exception as e:
|
||
print(f"❌ 错误: {e}", file=sys.stderr)
|
||
sys.exit(1)
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|