xiaozhi-sdk 0.2.5__py3-none-any.whl → 0.2.6__py3-none-any.whl

xiaozhi_sdk/__init__.py

@@ -1,3 +1,3 @@
-__version__ = "0.2.5"
+__version__ = "0.2.6"
 
 from xiaozhi_sdk.core import XiaoZhiWebsocket  # noqa

xiaozhi_sdk/cli.py

@@ -2,6 +2,7 @@ import asyncio
 import io
 import logging
 import time
+import uuid
 from collections import deque
 from typing import Optional
 
@@ -12,7 +13,11 @@ import sounddevice as sd
 from PIL import ImageGrab
 
 from xiaozhi_sdk import XiaoZhiWebsocket
-from xiaozhi_sdk.config import INPUT_SERVER_AUDIO_SAMPLE_RATE
+from xiaozhi_sdk.config import (
+    INPUT_SERVER_AUDIO_CHANNELS,
+    INPUT_SERVER_AUDIO_FRAME_DURATION,
+    INPUT_SERVER_AUDIO_SAMPLE_RATE,
+)
 
 # 定义自定义日志级别
 INFO1 = 21
@@ -73,6 +78,7 @@ logger.setLevel(logging.DEBUG)
 input_audio_buffer: deque[bytes] = deque()
 is_playing_audio = False
 is_end = False
+human_speak_time = None
 
 
 def get_image_byte(data):
@@ -103,14 +109,25 @@ def get_image_byte(data):
 
 
 async def handle_message(message):
+    global is_playing_audio
+    global human_speak_time
+
     """处理接收到的消息"""
     global is_end
-    if message["type"] == "stt":  # 人类语音
+
+    if message["type"] == "tts" and message["state"] == "start":  # start
+        pass
+
+    elif message["type"] == "stt":  # 人类语音
+        human_speak_time = time.time()
         logger.info1("human: %s", message["text"])
+
     elif message["type"] == "tts" and message["state"] == "sentence_start":  # AI语音
+        is_playing_audio = True  # 防止打断
         logger.info2("AI: %s", message["text"])
+
     elif message["type"] == "tts" and message["state"] == "stop":
-        pass
+        is_playing_audio = False
         # logger.info2("播放结束")
         # logger.info("聆听中...")
     elif message["type"] == "llm":  # 表情
@@ -123,13 +140,14 @@ async def handle_message(message):
         is_end = True
 
 
-async def play_assistant_audio(audio_queue: deque[bytes], enable_audio):
+async def play_assistant_audio(audio_queue: deque[bytes], enable_audio, audio_samplerate):
     """播放音频流"""
     global is_playing_audio
+    global human_speak_time
 
     stream = None
     if enable_audio:
-        stream = sd.OutputStream(samplerate=INPUT_SERVER_AUDIO_SAMPLE_RATE, channels=1, dtype=np.int16)
+        stream = sd.OutputStream(samplerate=audio_samplerate, channels=INPUT_SERVER_AUDIO_CHANNELS, dtype=np.int16)
         stream.start()
     last_audio_time = None
 
@@ -144,6 +162,10 @@ async def play_assistant_audio(audio_queue: deque[bytes], enable_audio):
             continue
 
         is_playing_audio = True
+        if human_speak_time:
+            logger.debug("首个音频包响应时间：%s 秒", time.time() - human_speak_time)
+            human_speak_time = None
+
         pcm_data = audio_queue.popleft()
         if stream:
             stream.write(pcm_data)
@@ -165,10 +187,16 @@ class XiaoZhiClient:
         self.mac_address = ""
         self.wake_word = wake_word
 
-    async def start(self, mac_address: str, serial_number: str, license_key: str, enable_audio):
+    async def start(self, mac_address: str, serial_number: str, license_key: str, enable_audio, audio_samplerate):
         """启动客户端连接"""
         self.mac_address = mac_address
-        self.xiaozhi = XiaoZhiWebsocket(handle_message, url=self.url, ota_url=self.ota_url, wake_word=self.wake_word)
+        self.xiaozhi = XiaoZhiWebsocket(
+            handle_message,
+            url=self.url,
+            ota_url=self.ota_url,
+            wake_word=self.wake_word,
+            audio_sample_rate=audio_samplerate,
+        )
         from xiaozhi_sdk.utils.mcp_tool import take_photo
 
         take_photo["tool_func"] = get_image_byte
@@ -178,7 +206,7 @@ class XiaoZhiClient:
             self.mac_address, aec=False, serial_number=serial_number, license_key=license_key
         )
 
-        asyncio.create_task(play_assistant_audio(self.xiaozhi.output_audio_queue, enable_audio))
+        asyncio.create_task(play_assistant_audio(self.xiaozhi.output_audio_queue, enable_audio, audio_samplerate))
 
     def audio_callback(self, indata, frames, time, status):
         """音频输入回调函数"""
@@ -202,30 +230,60 @@ class XiaoZhiClient:
 
 
 async def run_client(
-    mac_address: str, url: str, ota_url: str, serial_number: str, license_key: str, enable_audio: bool, wake_word: str
+    mac_address: str,
+    url: str,
+    ota_url: str,
+    serial_number: str,
+    license_key: str,
+    enable_audio: bool,
+    wake_word: str,
+    audio_samplerate: int,
 ):
     """运行客户端的异步函数"""
     logger.debug("Recording... Press Ctrl+C to stop.")
     client = XiaoZhiClient(url, ota_url, wake_word)
-    await client.start(mac_address, serial_number, license_key, enable_audio)
-
-    with sd.InputStream(callback=client.audio_callback, channels=1, samplerate=16000, blocksize=960):
+    await client.start(mac_address, serial_number, license_key, enable_audio, audio_samplerate)
+    blocksize = audio_samplerate * INPUT_SERVER_AUDIO_FRAME_DURATION // 1000
+    with sd.InputStream(
+        callback=client.audio_callback,
+        channels=INPUT_SERVER_AUDIO_CHANNELS,
+        samplerate=audio_samplerate,
+        blocksize=blocksize,
+    ):
+        logger.info("聆听中...")
         await client.process_audio_input()
 
 
+def get_mac_address():
+    mac = uuid.getnode()
+    mac_addr = ":".join(["%02x" % ((mac >> ele) & 0xFF) for ele in range(40, -8, -8)])
+    return mac_addr
+
+
 @click.command()
-@click.argument("mac_address")
+@click.argument("mac_address", required=False)
 @click.option("--url", help="服务端websocket地址")
 @click.option("--ota_url", help="OTA地址")
 @click.option("--serial_number", default="", help="设备的序列号")
 @click.option("--license_key", default="", help="设备的授权密钥")
 @click.option("--enable_audio", default=True, help="是否开启音频播放")
 @click.option("--wake_word", default="", help="唤醒词")
+@click.option("--audio_samplerate", default=INPUT_SERVER_AUDIO_SAMPLE_RATE, help="音频采样率，默认: 24000")
 def main(
-    mac_address: str, url: str, ota_url: str, serial_number: str, license_key: str, enable_audio: bool, wake_word: str
+    mac_address: str,
+    url: str,
+    ota_url: str,
+    serial_number: str,
+    license_key: str,
+    enable_audio: bool,
+    wake_word: str,
+    audio_samplerate: int,
 ):
     """小智SDK客户端
 
     MAC_ADDRESS: 设备的MAC地址 (格式: XX:XX:XX:XX:XX:XX)
     """
-    asyncio.run(run_client(mac_address, url, ota_url, serial_number, license_key, enable_audio, wake_word))
+    mac_address = mac_address or get_mac_address()
+    asyncio.run(
+        run_client(mac_address, url, ota_url, serial_number, license_key, enable_audio, wake_word, audio_samplerate)
+    )

xiaozhi_sdk/config.py

@@ -1,3 +1,5 @@
-INPUT_SERVER_AUDIO_SAMPLE_RATE = 16000
+INPUT_SERVER_AUDIO_SAMPLE_RATE = 24000
+INPUT_SERVER_AUDIO_CHANNELS = 1
+INPUT_SERVER_AUDIO_FRAME_DURATION = 60
 
 OTA_URL = "https://api.tenclass.net/xiaozhi/ota"

xiaozhi_sdk/core.py

@@ -9,7 +9,11 @@ from typing import Any, Callable, Deque, Dict, Optional
 
 import websockets
 
-from xiaozhi_sdk.config import INPUT_SERVER_AUDIO_SAMPLE_RATE
+from xiaozhi_sdk.config import (
+    INPUT_SERVER_AUDIO_CHANNELS,
+    INPUT_SERVER_AUDIO_FRAME_DURATION,
+    INPUT_SERVER_AUDIO_SAMPLE_RATE,
+)
 from xiaozhi_sdk.iot import OtaDevice
 from xiaozhi_sdk.mcp import McpTool
 from xiaozhi_sdk.utils import get_wav_info, read_audio_file, setup_opus
@@ -27,15 +31,17 @@ class XiaoZhiWebsocket(McpTool):
         message_handler_callback: Optional[Callable] = None,
         url: Optional[str] = None,
         ota_url: Optional[str] = None,
-        audio_sample_rate: int = 16000,
-        audio_channels: int = 1,
+        audio_sample_rate: int = INPUT_SERVER_AUDIO_SAMPLE_RATE,
+        audio_channels: int = INPUT_SERVER_AUDIO_CHANNELS,
+        audio_frame_duration=INPUT_SERVER_AUDIO_FRAME_DURATION,
         wake_word: str = "",
     ):
         super().__init__()
         self.url = url
         self.ota_url = ota_url
         self.audio_channels = audio_channels
-        self.audio_opus = AudioOpus(audio_sample_rate, audio_channels)
+        self.audio_frame_duration = audio_frame_duration
+        self.audio_opus = AudioOpus(audio_sample_rate, audio_channels, audio_frame_duration)
         self.wake_word = wake_word
 
         # 客户端标识
@@ -70,13 +76,13 @@ class XiaoZhiWebsocket(McpTool):
         hello_message = {
             "type": "hello",
             "version": 1,
-            "features": {"mcp": True, "aec": aec},
+            "features": {"mcp": True, "aec": aec, "consistent_sample_rate": True},
             "transport": "websocket",
             "audio_params": {
                 "format": "opus",
-                "sample_rate": 16000,
-                "channels": 1,
-                "frame_duration": 60,
+                "sample_rate": self.audio_opus.sample_rate,
+                "channels": self.audio_opus.channels,
+                "frame_duration": self.audio_opus.frame_duration,
             },
         }
         await self.websocket.send(json.dumps(hello_message))
@@ -113,7 +119,7 @@ class XiaoZhiWebsocket(McpTool):
         current_dir = os.path.dirname(os.path.abspath(__file__))
         wav_path = os.path.join(current_dir, "../file/audio/greet.wav")
         framerate, channels = get_wav_info(wav_path)
-        audio_opus = AudioOpus(framerate, channels)
+        audio_opus = AudioOpus(framerate, channels, self.audio_frame_duration)
 
         for pcm_data in read_audio_file(wav_path):
             opus_data = await audio_opus.pcm_to_opus(pcm_data)
@@ -137,8 +143,8 @@ class XiaoZhiWebsocket(McpTool):
 
     async def send_silence_audio(self, duration_seconds: float = 1.2) -> None:
         """发送静音音频"""
-        frames_count = int(duration_seconds * 1000 / 60)
-        pcm_frame = b"\x00\x00" * int(INPUT_SERVER_AUDIO_SAMPLE_RATE / 1000 * 60)
+        frames_count = int(duration_seconds * 1000 / self.audio_opus.frame_duration)
+        pcm_frame = b"\x00\x00" * int(self.audio_opus.sample_rate / 1000 * self.audio_opus.frame_duration)
 
         for _ in range(frames_count):
             await self.send_audio(pcm_frame)
@@ -159,6 +165,7 @@ class XiaoZhiWebsocket(McpTool):
         data = json.loads(message)
         message_type = data["type"]
         if message_type == "hello":
+            self.audio_opus.set_out_audio_frame(data["audio_params"])
             self.hello_received.set()
             self.session_id = data["session_id"]
             return
@@ -219,7 +226,7 @@ class XiaoZhiWebsocket(McpTool):
 
         await self._send_hello(self.aec)
         await self._start_listen()
-        logger.debug("[websocket] Connection successful")
+        logger.debug("[websocket] Connection successful. mac_addr: %s", self.mac_addr)
         await asyncio.sleep(0.5)
 
     async def init_connection(
@@ -250,7 +257,9 @@ class XiaoZhiWebsocket(McpTool):
 
         if not await self.is_activate(ota_info):
             self.iot_task = asyncio.create_task(self._activate_iot_device(license_key, ota_info))
+            await self.send_wake_word("hi")
             logger.debug("[IOT] 设备未激活")
+            return
 
         if self.wake_word:
             await self.send_wake_word(self.wake_word)

xiaozhi_sdk/opus.py

@@ -1,61 +1,59 @@
-import math
-
 import av
 import numpy as np
 import opuslib
 
-from xiaozhi_sdk.config import INPUT_SERVER_AUDIO_SAMPLE_RATE
-
 
 class AudioOpus:
 
-    def __init__(self, sample_rate, channels):
+    def __init__(self, sample_rate, channels, frame_duration):
+        self.frame_duration = frame_duration
         self.sample_rate = sample_rate
         self.channels = channels
+        self.frame_size = self.sample_rate * self.frame_duration // 1000
 
         # 创建 Opus 编码器
-        self.opus_encoder = opuslib.Encoder(
-            fs=sample_rate, channels=channels, application=opuslib.APPLICATION_VOIP  # 采样率  # 单声道  # 语音应用
-        )
+        self.opus_encoder = opuslib.Encoder(fs=sample_rate, channels=channels, application=opuslib.APPLICATION_VOIP)
+
+        self.resampler = av.AudioResampler(format="s16", layout="mono", rate=sample_rate)
+
+    def set_out_audio_frame(self, audio_params):
+        # 小智服务端 的 音频信息
+        self.out_frame_size = audio_params["sample_rate"] * audio_params["frame_duration"] // 1000
 
         # 创建 Opus 解码器
         self.opus_decoder = opuslib.Decoder(
-            fs=INPUT_SERVER_AUDIO_SAMPLE_RATE,  # 采样率
-            channels=1,  # 单声道
+            fs=audio_params["sample_rate"],  # 采样率
+            channels=audio_params["channels"],  # 单声道
         )
 
-        self.resampler = av.AudioResampler(format="s16", layout="mono", rate=sample_rate)
-
     async def pcm_to_opus(self, pcm):
         pcm_array = np.frombuffer(pcm, dtype=np.int16)
         pcm_bytes = pcm_array.tobytes()
-        return self.opus_encoder.encode(pcm_bytes, 960)
-
-    @staticmethod
-    def to_n_960(samples) -> np.ndarray:
-        n = math.ceil(samples.shape[0] / 960)
-        arr_padded = np.pad(samples, (0, 960 * n - samples.shape[0]), mode="constant", constant_values=0)
-        return arr_padded.reshape(n, 960)
+        return self.opus_encoder.encode(pcm_bytes, self.frame_size)
 
     async def change_sample_rate(self, pcm_array) -> np.ndarray:
-        if self.sample_rate == INPUT_SERVER_AUDIO_SAMPLE_RATE:
-            return self.to_n_960(pcm_array)
-
+        # 采样率 变更
         frame = av.AudioFrame.from_ndarray(np.array(pcm_array).reshape(1, -1), format="s16", layout="mono")
-        frame.sample_rate = INPUT_SERVER_AUDIO_SAMPLE_RATE  # Assuming input is 16kHz
+        frame.sample_rate = self.opus_decoder._fs
         resampled_frames = self.resampler.resample(frame)
         samples = resampled_frames[0].to_ndarray().flatten()
         new_frame = av.AudioFrame.from_ndarray(
             samples.reshape(1, -1),
             format="s16",
             layout="mono",
+            # layout="stereo",
         )
         new_frame.sample_rate = self.sample_rate
         new_samples = new_frame.to_ndarray().flatten()
-        return self.to_n_960(new_samples)
+
+        # 不足 self.frame_size 补 0
+        samples_padded = np.pad(
+            new_samples, (0, self.frame_size - new_samples.size), mode="constant", constant_values=0
+        )
+        return samples_padded.reshape(1, self.frame_size)
 
     async def opus_to_pcm(self, opus) -> np.ndarray:
-        pcm_data = self.opus_decoder.decode(opus, 960)
+        pcm_data = self.opus_decoder.decode(opus, frame_size=self.out_frame_size)
         pcm_array = np.frombuffer(pcm_data, dtype=np.int16)
         samples = await self.change_sample_rate(pcm_array)
         return samples

{xiaozhi_sdk-0.2.5.dist-info → xiaozhi_sdk-0.2.6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xiaozhi-sdk
-Version: 0.2.5
+Version: 0.2.6
 Summary: 一个用于连接和控制小智智能设备的Python SDK，支持实时音频通信、MCP工具集成和设备管理功能。
 Author-email: dairoot <623815825@qq.com>
 License-Expression: MIT
@@ -63,6 +63,10 @@ python -m xiaozhi_sdk --help
 #### 连接设备（需要提供 MAC 地址）
 
 ```bash
+# 默认本机 mac 地址
+python -m xiaozhi_sdk
+
+# 指定 mac 地址
 python -m xiaozhi_sdk 00:22:44:66:88:00
 ```
 

{xiaozhi_sdk-0.2.5.dist-info → xiaozhi_sdk-0.2.6.dist-info}/RECORD

@@ -8,19 +8,19 @@ file/opus/linux-x64-libopus.so,sha256=FmXJqkxLpDzNFOHYkmOzmsp1hP0eIS5b6x_XfOs-IQ
 file/opus/macos-arm64-libopus.dylib,sha256=H7wXwkrGwb-hesMMZGFxWb0Ri1Y4m5GWiKsd8CfOhE8,357584
 file/opus/macos-x64-libopus.dylib,sha256=MqyL_OjwSACF4Xs_-KrGbcScy4IEprr5Rlkk3ddZye8,550856
 file/opus/windows-opus.dll,sha256=kLfhioMvbJhOgNMAldpWk3DCZqC5Xd70LRbHnACvAnw,463360
-xiaozhi_sdk/__init__.py,sha256=VMDxwPy8TInmElVlOBSsiIjvXCQ7OCLzRJosuhXY8us,77
+xiaozhi_sdk/__init__.py,sha256=hszHgr7Kf8Xfmvits8MjRxs-JmoGZq1Xc6fL6ChJipg,77
 xiaozhi_sdk/__main__.py,sha256=i0ZJdHUqAKg9vwZrK_w0TJkzdotTYTK8aUeSPcJc1ks,210
-xiaozhi_sdk/cli.py,sha256=CEpjXW-vnT8DieaK1MV0AYCG4bmK2_vpIz-9zhWHVgE,7001
-xiaozhi_sdk/config.py,sha256=h4mpMeBf2vT9qYAqCCbGVGmMemkgk98pcXP2Rh4TEFc,89
-xiaozhi_sdk/core.py,sha256=JfYH7dyKxb1ikVkam74V_SMvn07MTdBfY7u3X6OdUxk,10723
+xiaozhi_sdk/cli.py,sha256=7qL3-y_h5M00og2iqdaGKZx4-z5M3TKlhBAz1KeWBAY,8427
+xiaozhi_sdk/config.py,sha256=tIQMBEoCkHXS-lR2IUSFNRkUEJ6EHF2rc_stx0gEIZU,160
+xiaozhi_sdk/core.py,sha256=bpGInL2mPK-Y_n4IfJmSdeXnpAa9X2rqKJLxIgE8pNk,11345
 xiaozhi_sdk/iot.py,sha256=VVAheynp1iV4GCaoPywQWpKtlyoACDLswH8yfV_JZgI,2699
 xiaozhi_sdk/mcp.py,sha256=g80dijPgWQ-npx7sD4it4_kUMCTO40gIcbXoLgF1ksE,6712
-xiaozhi_sdk/opus.py,sha256=r3nnYg0ZKAJTreb_3nKgfHJh06MJiMvnNMPO1SWdoMM,2224
+xiaozhi_sdk/opus.py,sha256=f0nus9vxNo7TOfVEvrbJbpggWRCQlK2UubJF_Nu0Ups,2329
 xiaozhi_sdk/utils/__init__.py,sha256=XKSHWoFmuSkpwaIr308HybRzfFIXoT1Fd-eUKo_im6Y,1705
 xiaozhi_sdk/utils/mcp_tool.py,sha256=T6OIrSqcyAHQ85sduz5Klx646SoEnGD5ROBTKoX6NhE,4207
 xiaozhi_sdk/utils/tool_func.py,sha256=imwehfUlENjelYmGbGYgb6C82-ijs53XCxrtCpqrJps,3152
-xiaozhi_sdk-0.2.5.dist-info/licenses/LICENSE,sha256=Vwgps1iODKl43cAtME_0dawTjAzNW-O2BWiN5BHggww,1085
-xiaozhi_sdk-0.2.5.dist-info/METADATA,sha256=h55NbddO9BM2IwSxhRRSJi_X40dmZHn-eItvi9Gs2bw,2121
-xiaozhi_sdk-0.2.5.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-xiaozhi_sdk-0.2.5.dist-info/top_level.txt,sha256=nBpue4hU5Ykm5CtYPsAdxSa_yqbtZsIT_gF_EkBaJPM,12
-xiaozhi_sdk-0.2.5.dist-info/RECORD,,
+xiaozhi_sdk-0.2.6.dist-info/licenses/LICENSE,sha256=Vwgps1iODKl43cAtME_0dawTjAzNW-O2BWiN5BHggww,1085
+xiaozhi_sdk-0.2.6.dist-info/METADATA,sha256=vgOY2f3vdRtAtlsLiErTlOHXlNS-yd2HetBAONAa72g,2190
+xiaozhi_sdk-0.2.6.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+xiaozhi_sdk-0.2.6.dist-info/top_level.txt,sha256=nBpue4hU5Ykm5CtYPsAdxSa_yqbtZsIT_gF_EkBaJPM,12
+xiaozhi_sdk-0.2.6.dist-info/RECORD,,