add vector server, test=doc

demos/speaker_verification/README.md

@@ -14,7 +14,7 @@ see [installation](https://github.com/PaddlePaddle/PaddleSpeech/blob/develop/doc
 You can choose one way from easy, meduim and hard to install paddlespeech.
 
 ### 2. Prepare Input File
-The input of this demo should be a WAV file(`.wav`), and the sample rate must be the same as the model.
+The input of this cli demo should be a WAV file(`.wav`), and the sample rate must be the same as the model.
 
 Here are sample files for this demo that can be downloaded:
 ```bash

demos/speaker_verification/README_cn.md

@@ -4,16 +4,16 @@
 ## 介绍
 声纹识别是一项用计算机程序自动提取说话人特征的技术。
 
-这个 demo 是一个从给定音频文件提取说话人特征，它可以通过使用 `PaddleSpeech` 的单个命令或 python 中的几行代码来实现。
+这个 demo 是从一个给定音频文件中提取说话人特征，它可以通过使用 `PaddleSpeech` 的单个命令或 python 中的几行代码来实现。
 
 ## 使用方法
 ### 1. 安装
 请看[安装文档](https://github.com/PaddlePaddle/PaddleSpeech/blob/develop/docs/source/install_cn.md)。
 
-你可以从 easy，medium，hard 三中方式中选择一种方式安装。
+你可以从easy medium，hard 三种方式中选择一种方式安装。
 
 ### 2. 准备输入
-这个 demo 的输入应该是一个 WAV 文件（`.wav`），并且采样率必须与模型的采样率相同。
+声纹cli demo 的输入应该是一个 WAV 文件（`.wav`），并且采样率必须与模型的采样率相同。
 
 可以下载此 demo 的示例音频：
 ```bash

demos/streaming_asr_server/websocket_client.py

@@ -28,6 +28,7 @@ def main(args):
     handler = ASRWsAudioHandler(
         args.server_ip,
         args.port,
+        endpoint=args.endpoint,
         punc_server_ip=args.punc_server_ip,
         punc_server_port=args.punc_server_port)
     loop = asyncio.get_event_loop()
@@ -36,7 +37,7 @@ def main(args):
     if args.wavfile and os.path.exists(args.wavfile):
         logger.info(f"start to process the wavscp: {args.wavfile}")
         result = loop.run_until_complete(handler.run(args.wavfile))
-        result = result["result"]
+        # result = result["result"]
         logger.info(f"asr websocket client finished : {result}")
 
     # support to process batch audios from wav.scp 
@@ -69,7 +70,11 @@ def main(args):
         default=8091,
         dest="punc_server_port",
         help='Punctuation server port')
-
+    parser.add_argument(
+        "--endpoint",
+        type=str,
+        default="/paddlespeech/asr/streaming",
+        help="ASR websocket endpoint")
     parser.add_argument(
         "--wavfile",
         action="store",

paddlespeech/cli/vector/infer.py

@@ -272,7 +272,8 @@ def _init_from_path(self,
                         model_type: str='ecapatdnn_voxceleb12',
                         sample_rate: int=16000,
                         cfg_path: Optional[os.PathLike]=None,
-                        ckpt_path: Optional[os.PathLike]=None):
+                        ckpt_path: Optional[os.PathLike]=None,
+                        task=None):
         """Init the neural network from the model path
 
         Args:

paddlespeech/server/README_cn.md

@@ -63,3 +63,23 @@ paddlespeech_server start --config_file conf/tts_online_application.yaml
 ```
 paddlespeech_client tts_online  --server_ip 127.0.0.1 --port 8092 --input "您好，欢迎使用百度飞桨深度学习框架！" --output output.wav
 ```
+
+## 声纹识别
+
+### 启动声纹识别服务
+
+```
+paddlespeech_server start --config_file conf/vector_application.yaml
+```
+
+### 获取说话人音频声纹
+
+```
+paddlespeech_client vector --task spk  --server_ip 127.0.0.1 --port 8090 --input 85236145389.wav
+```
+
+### 两个说话人音频声纹打分
+
+```
+paddlespeech_client vector --task score  --server_ip 127.0.0.1 --port 8090 --enroll 123456789.wav --test 85236145389.wav
+```

paddlespeech/server/bin/paddlespeech_client.py

@@ -35,7 +35,7 @@
 
 __all__ = [
     'TTSClientExecutor', 'TTSOnlineClientExecutor', 'ASRClientExecutor',
-    'ASROnlineClientExecutor', 'CLSClientExecutor'
+    'ASROnlineClientExecutor', 'CLSClientExecutor', 'VectorClientExecutor'
 ]
 
 
@@ -583,3 +583,100 @@ def __call__(self, input: str, server_ip: str="127.0.0.1", port: int=8090):
         response_dict = res.json()
         punc_text = response_dict["result"]["punc_text"]
         return punc_text
+
+
+@cli_client_register(
+    name='paddlespeech_client.vector', description='visit the vector service')
+class VectorClientExecutor(BaseExecutor):
+    def __init__(self):
+        super(VectorClientExecutor, self).__init__()
+        self.parser = argparse.ArgumentParser(
+            prog='paddlespeech_client.vector', add_help=True)
+        self.parser.add_argument(
+            '--server_ip', type=str, default='127.0.0.1', help='server ip')
+        self.parser.add_argument(
+            '--port', type=int, default=8090, help='server port')
+        self.parser.add_argument(
+            '--input',
+            type=str,
+            default=None,
+            help='sentence to be process by text server.')
+        self.parser.add_argument(
+            '--task', type=str, default="spk", help="The vector service task")
+        self.parser.add_argument(
+            "--enroll", type=str, default=None, help="The enroll audio")
+        self.parser.add_argument(
+            "--test", type=str, default=None, help="The test audio")
+
+    def execute(self, argv: List[str]) -> bool:
+        """Execute the request from the argv.
+
+        Args:
+            argv (List): the request arguments
+
+        Returns:
+            str: the request flag
+        """
+        args = self.parser.parse_args(argv)
+        input_ = args.input
+        server_ip = args.server_ip
+        port = args.port
+        task = args.task
+
+        try:
+            time_start = time.time()
+            res = self(
+                input=input_,
+                server_ip=server_ip,
+                port=port,
+                enroll_audio=args.enroll,
+                test_audio=args.test,
+                task=task)
+            time_end = time.time()
+            logger.info(f"The vector: {res}")
+            logger.info("Response time %f s." % (time_end - time_start))
+            return True
+        except Exception as e:
+            logger.error("Failed to extract vector.")
+            logger.error(e)
+            return False
+
+    @stats_wrapper
+    def __call__(self,
+                 input: str,
+                 server_ip: str="127.0.0.1",
+                 port: int=8090,
+                 audio_format: str="wav",
+                 sample_rate: int=16000,
+                 enroll_audio: str=None,
+                 test_audio: str=None,
+                 task="spk"):
+        """
+        Python API to call text executor.
+
+        Args:
+            input (str): the request sentence text
+            server_ip (str, optional): the server ip. Defaults to "127.0.0.1".
+            port (int, optional): the server port. Defaults to 8090.
+
+        Returns:
+            str: the punctuation text
+        """
+        if task == "spk":
+            from paddlespeech.server.utils.audio_handler import VectorHttpHandler
+            logger.info("vector http client start")
+            logger.info(f"the input audio: {input}")
+            handler = VectorHttpHandler(server_ip=server_ip, port=port)
+            res = handler.run(input, audio_format, sample_rate)
+            return res
+        elif task == "score":
+            from paddlespeech.server.utils.audio_handler import VectorScoreHttpHandler
+            logger.info("vector score http client start")
+            logger.info(
+                f"enroll audio: {enroll_audio}, test audio: {test_audio}")
+            handler = VectorScoreHttpHandler(server_ip=server_ip, port=port)
+            res = handler.run(enroll_audio, test_audio, audio_format,
+                              sample_rate)
+            logger.info(f"The vector score is: {res}")
+        else:
+            logger.error(f"Sorry, we have not support such task {task}")

paddlespeech/server/conf/application.yaml

@@ -11,7 +11,7 @@ port: 8090
 # protocol = ['websocket', 'http'] (only one can be selected). 
 # http only support offline engine type.
 protocol: 'http'
-engine_list: ['asr_python', 'tts_python', 'cls_python', 'text_python']
+engine_list: ['asr_python', 'tts_python', 'cls_python', 'text_python', 'vector_python']
 
 
 #################################################################################
@@ -166,4 +166,15 @@ text_python:
     cfg_path: # [optional]
     ckpt_path: # [optional]
     vocab_file: # [optional]
+    device:  # set 'gpu:id' or 'cpu'
+
+
+################################### Vector ######################################
+################### Vector task: spk; engine_type: python #######################
+vector_python:
+    task: spk
+    model_type: 'ecapatdnn_voxceleb12'
+    sample_rate: 16000
+    cfg_path: # [optional]
+    ckpt_path: # [optional]
     device:  # set 'gpu:id' or 'cpu'

paddlespeech/server/conf/vector_application.yaml

@@ -0,0 +1,32 @@
+# This is the parameter configuration file for PaddleSpeech Serving.
+
+#################################################################################
+#                             SERVER SETTING                                    #
+#################################################################################
+host: 0.0.0.0
+port: 8090
+
+# The task format in the engin_list is: <speech task>_<engine type>
+# protocol = ['http'] (only one can be selected). 
+# http only support offline engine type.
+protocol: 'http'
+engine_list: ['vector_python']
+
+
+#################################################################################
+#                                ENGINE CONFIG                                  #
+#################################################################################
+
+################################### Vector ######################################
+################### Vector task: spk; engine_type: python #######################
+vector_python:
+    task: spk
+    model_type: 'ecapatdnn_voxceleb12'
+    sample_rate: 16000
+    cfg_path: # [optional]
+    ckpt_path: # [optional]
+    device: # set 'gpu:id' or 'cpu'
+
+
+
+

paddlespeech/server/engine/asr/online/asr_engine.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 import copy
 import os
+import time
 from typing import Optional
 
 import numpy as np
@@ -153,6 +154,12 @@ def init(self):
             self.n_shift = self.preprocess_conf.process[0]['n_shift']
 
     def extract_feat(self, samples):
+
+        # we compute the elapsed time of first char occuring 
+        # and we record the start time at the first pcm sample arraving
+        # if self.first_char_occur_elapsed is not None:
+        #     self.first_char_occur_elapsed = time.time()
+
         if "deepspeech2online" in self.model_type:
             # self.reamined_wav stores all the samples, 
             # include the original remained_wav and this package samples
@@ -290,6 +297,7 @@ def reset(self):
         self.chunk_num = 0
         self.global_frame_offset = 0
         self.result_transcripts = ['']
+        self.first_char_occur_elapsed = None
 
     def decode(self, is_finished=False):
         if "deepspeech2online" in self.model_type:

paddlespeech/server/engine/engine_factory.py

@@ -49,5 +49,8 @@ def get_engine(engine_name: Text, engine_type: Text):
         elif engine_name.lower() == 'text' and engine_type.lower() == 'python':
             from paddlespeech.server.engine.text.python.text_engine import TextEngine
             return TextEngine()
+        elif engine_name.lower() == 'vector' and engine_type.lower() == 'python':
+            from paddlespeech.server.engine.vector.python.vector_engine import VectorEngine
+            return VectorEngine()
         else:
             return None