Merge pull request #1915 from YourKnight1/main

调整PaddleSpeechTTS流式传输
2026-07-22 07:03:53 +08:00 · 2025-08-07 23:42:16 +08:00
parent 6091eae7fc 04746961f2
commit b3f603732e
2 changed files with 167 additions and 0 deletions
@@ -885,3 +885,18 @@ TTS:
    access_token: "U4YdYXVfpwWnk2t5Gp822zWPCuORyeJL"
    voice: "OUeAo1mhq6IBExi"
    output_dir: tmp/
+  PaddleSpeechTTS:
+    #百度飞浆 PaddleSpeech 支持本地离线部署 支持模型训练
+    #框架地址 https://www.paddlepaddle.org.cn/
+    #项目地址 https://github.com/PaddlePaddle/PaddleSpeech
+    #SpeechServerDemo https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/demos/speech_server
+    #流式传输请参考 https://github.com/PaddlePaddle/PaddleSpeech/wiki/PaddleSpeech-Server-WebSocket-API
+    type: paddle_speech
+    protocol: websocket # protocol choices = ['websocket', 'http']
+    url: ws://127.0.0.1:8092/paddlespeech/tts/streaming  # TTS 服务的 URL 地址，指向本地服务器 [websocket默认ws://127.0.0.1:8092/paddlespeech/tts/streaming，http默认http://127.0.0.1:8090/paddlespeech/tts]
+    spk_id: 0  # 发音人 ID，0 通常表示默认的发音人
+    sample_rate: 24000  # 采样率 [websocket默认24000，http默认0 自动选择]
+    speed: 1.0  # 语速，1.0 表示正常语速，>1 表示加快，<1 表示减慢
+    volume: 1.0  # 音量，1.0 表示正常音量，>1 表示增大，<1 表示减小
+    save_path: ./streaming_tts.wav  # 服务器生成的语音文件保存路径
+
@@ -0,0 +1,152 @@
+import asyncio
+import json
+import base64
+import aiohttp
+import numpy as np
+import io
+import wave
+import websockets
+from core.providers.tts.base import TTSProviderBase
+from config.logger import setup_logging
+
+TAG = __name__
+logger = setup_logging()
+
+
+class TTSProvider(TTSProviderBase):
+    def __init__(self, config, delete_audio_file):
+        super().__init__(config, delete_audio_file)
+        self.url = config.get("url", "ws://192.168.1.10:8092/paddlespeech/tts/streaming")
+        self.protocol = config.get("protocol", "websocket")
+        self.spk_id = config.get("spk_id", 0)
+        self.sample_rate = config.get("sample_rate", 24000)
+        self.speed = config.get("speed", 1.0)
+        self.volume = config.get("volume", 1.0)
+        self.save_path = config.get("save_path", "./streaming_tts.wav")
+
+    async def pcm_to_wav(self, pcm_data: bytes, sample_rate: int = 24000, num_channels: int = 1,
+                         bits_per_sample: int = 16) -> bytes:
+        """
+        将 PCM 数据转换为 WAV 文件并返回字节数据
+        :param pcm_data: PCM 数据（原始字节流）
+        :param sample_rate: 音频采样率，默认为24000
+        :param num_channels: 声道数，默认为单声道
+        :param bits_per_sample: 每个样本的位数，默认为16
+        :return: WAV 格式的字节数据
+        """
+        byte_data = np.frombuffer(pcm_data, dtype=np.int16)  # 16位PCM
+        wav_io = io.BytesIO()
+
+        with wave.open(wav_io, "wb") as wav_file:
+            wav_file.setnchannels(num_channels)
+            wav_file.setsampwidth(bits_per_sample // 8)
+            wav_file.setframerate(sample_rate)
+            wav_file.writeframes(byte_data.tobytes())
+
+        return wav_io.getvalue()
+
+    async def text_to_speak(self, text, output_file):
+        if self.protocol == "websocket":
+            return await self.text_streaming(text, output_file)
+        elif self.protocol == "http":
+            return await self.text(text, output_file)
+        else:
+            raise ValueError("Unsupported protocol. Please use 'websocket' or 'http'.")
+
+    async def text(self, text, output_file):
+        request_json = {
+            "text": text,
+            "spk_id": self.spk_id,
+            "speed": self.speed,
+            "volume": self.volume,
+            "sample_rate": self.sample_rate,
+            "save_path": self.save_path
+        }
+
+        try:
+            async with aiohttp.ClientSession() as session:
+                async with session.post(self.url, json=request_json) as resp:
+                    if resp.status == 200:
+                        resp_json = await resp.json()
+                        if resp_json.get("success"):
+                            data = resp_json["result"]
+                            audio_bytes = base64.b64decode(data["audio"])
+                            if output_file:
+                                with open(output_file, "wb") as file_to_save:
+                                    file_to_save.write(audio_bytes)
+                            else:
+                                return audio_bytes
+                        else:
+                            raise Exception(
+                                f"Error: {resp_json.get('message', 'Unknown error')} while processing text: {text}")
+                    else:
+                        raise Exception(
+                            f"HTTP Error: {resp.status} - {await resp.text()} while processing text: {text}")
+        except Exception as e:
+            raise Exception(f"Error during TTS HTTP request: {e} while processing text: {text}")
+
+    async def text_streaming(self, text, output_file):
+        try:
+            # 使用 websockets 异步连接到 WebSocket 服务器
+            async with websockets.connect(self.url) as ws:
+                # 发送开始请求
+                start_request = {
+                    "task": "tts",
+                    "signal": "start"
+                }
+                await ws.send(json.dumps(start_request))
+
+                # 接收开始响应并提取 session_id
+                start_response = await ws.recv()
+                start_response = json.loads(start_response)  # 解析 JSON 响应
+                if start_response.get("status") != 0:
+                    raise Exception(f"连接失败: {start_response.get('signal')}")
+
+                session_id = start_response.get("session")
+
+                # 发送待合成的文本数据
+                data_request = {
+                    "text": text,
+                    "spk_id": self.spk_id,
+                }
+                await ws.send(json.dumps(data_request))
+
+                audio_chunks = b""
+                timeout_seconds = 60  # 设置超时
+                try:
+                    while True:
+                        response = await asyncio.wait_for(ws.recv(), timeout=timeout_seconds)
+                        response = json.loads(response)  # 解析 JSON 响应
+                        status = response.get("status")
+
+                        if status == 2:  # 最后一个数据包
+                            break
+                        else:
+                            # 拼接音频数据（base64 编码的 PCM 数据）
+                            audio_chunks += base64.b64decode(response.get("audio"))
+                except asyncio.TimeoutError:
+                    raise Exception(f"WebSocket 超时：等待音频数据超过 {timeout_seconds} 秒")
+
+                # 将拼接后的 PCM 数据转换为 WAV 格式
+                wav_data = await self.pcm_to_wav(audio_chunks)
+
+                # 结束请求
+                end_request = {
+                    "task": "tts",
+                    "signal": "end",
+                    "session": session_id  # 会话 ID 必须与开始请求中的一致
+                }
+                await ws.send(json.dumps(end_request))
+
+                # 接收结束响应避免服务抛出异常
+                await ws.recv()
+
+                # 返回或保存音频数据
+                if output_file:
+                    with open(output_file, "wb") as file_to_save:
+                        file_to_save.write(wav_data)
+                else:
+                    return wav_data
+
+        except Exception as e:
+            raise Exception(f"Error during TTS WebSocket request: {e} while processing text: {text}")