Fix a bug of streaming_tts_server #3865

SuiYunsy · 2024-10-25T13:18:43Z

PR types

Bug fixes

PR changes

Others

Describe

Fix "Failed to get model related files" when using fastspeech2_csmsc_onnx model in streaming_tts_server.

CLAassistant · 2024-10-25T13:18:50Z

All committers have signed the CLA.

zxcd · 2024-10-30T11:55:05Z

Which environment are you running in? This issue should not occur under the recommended configuration.

SuiYunsy · 2024-10-30T16:44:50Z

Which environment are you running in? This issue should not occur under the recommended configuration.

环境：

最新代码自己构建的镜像，以及 paddlecloud/paddlespeech:develop-cpu-latest

配置：

将默认配置里的 am

PaddleSpeech/demos/streaming_tts_server/conf/tts_online_application.yaml

Line 65 in 6316514

am: 'fastspeech2_cnndecoder_csmsc_onnx'

改成另一个可用模型 fastspeech2_csmsc_onnx

启动 paddlespeech_server 后就会报错 Failed to get model related files

究其原因，选择 fastspeech2_csmsc_onnx 模型

PaddleSpeech/paddlespeech/resource/pretrained_models.py

Lines 1735 to 1744 in 6316514

    
           "fastspeech2_csmsc_onnx-zh": { 
        
               '1.0': { 
        
                   'url': 
        
                   'https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_csmsc_onnx_0.2.0.zip', 
        
                   'md5': 
        
                   'fd3ad38d83273ad51f0ea4f4abf3ab4e', 
        
                   'ckpt': 
        
                   'fastspeech2_csmsc.onnx', 
        
                   'phones_dict': 
        
                   'phone_id_map.txt',

这里的 ckpt 键对应的值并非列表，但是目前的代码中是这么处理的，就会报错：

PaddleSpeech/paddlespeech/server/engine/tts/online/onnx/tts_engine.py

Lines 70 to 81 in 6316514

    
           if am == "fastspeech2_csmsc_onnx": 
        
               # get model info 
        
               if am_ckpt is None or phones_dict is None: 
        
                   self.task_resource.set_task_model( 
        
                       model_tag=am_tag, 
        
                       model_type=0,  # am 
        
                       version=None,  # default version 
        
                   ) 
        
                   self.am_res_path = self.task_resource.res_dir 
        
                   self.am_ckpt = os.path.join( 
        
                       self.am_res_path, self.task_resource.res_dict['ckpt'][0]) 
        
                   # must have phones_dict in acoustic

相对地，默认配置选择的是 fastspeech2_cnndecoder_csmsc_onnx

PaddleSpeech/paddlespeech/resource/pretrained_models.py

Lines 1795 to 1805 in 6316514

    
           "fastspeech2_cnndecoder_csmsc_onnx-zh": { 
        
               '1.0': { 
        
                   'url': 
        
                   'https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_cnndecoder_csmsc_streaming_onnx_1.0.0.zip', 
        
                   'md5': 
        
                   '5f70e1a6bcd29d72d54e7931aa86f266', 
        
                   'ckpt': [ 
        
                       'fastspeech2_csmsc_am_encoder_infer.onnx', 
        
                       'fastspeech2_csmsc_am_decoder.onnx', 
        
                       'fastspeech2_csmsc_am_postnet.onnx', 
        
                   ],

这里的 ckpt 键对应的值为列表，代码中有正确处理，可以正常运行：

PaddleSpeech/paddlespeech/server/engine/tts/online/onnx/tts_engine.py

Lines 94 to 108 in 6316514

    
           elif am == "fastspeech2_cnndecoder_csmsc_onnx": 
        
               if am_ckpt is None or am_stat is None or phones_dict is None: 
        
                   self.task_resource.set_task_model( 
        
                       model_tag=am_tag, 
        
                       model_type=0,  # am 
        
                       version=None,  # default version 
        
                   ) 
        
                   self.am_res_path = self.task_resource.res_dir 
        
                   self.am_encoder_infer = os.path.join( 
        
                       self.am_res_path, self.task_resource.res_dict['ckpt'][0]) 
        
                   self.am_decoder = os.path.join( 
        
                       self.am_res_path, self.task_resource.res_dict['ckpt'][1]) 
        
                   self.am_postnet = os.path.join( 
        
                       self.am_res_path, self.task_resource.res_dict['ckpt'][2]) 
        
                   # must have phones_dict in acoustic

所以有了这个 PR😊
不足之处还望赐教😀

zxcd · 2024-10-31T06:30:16Z

好的，也麻烦把CI过一下，方便合入代码:D

SuiYunsy · 2024-10-31T14:59:46Z

好的，也麻烦把CI过一下，方便合入代码:D

想问问上图这个CodeStyle，它建议是改成这样：

但是我仅仅把 [0] 删掉了，文件中其他部分的代码也是这个格式：

PaddleSpeech/paddlespeech/server/engine/tts/online/onnx/tts_engine.py

Lines 78 to 84 in 6316514

    
           self.am_res_path = self.task_resource.res_dir 
        
           self.am_ckpt = os.path.join( 
        
               self.am_res_path, self.task_resource.res_dict['ckpt'][0]) 
        
           # must have phones_dict in acoustic 
        
           self.phones_dict = os.path.join( 
        
               self.am_res_path, 
        
               self.task_resource.res_dict['phones_dict'])

PaddleSpeech/paddlespeech/server/engine/tts/online/onnx/tts_engine.py

Lines 101 to 114 in 6316514

    
           self.am_res_path = self.task_resource.res_dir 
        
           self.am_encoder_infer = os.path.join( 
        
               self.am_res_path, self.task_resource.res_dict['ckpt'][0]) 
        
           self.am_decoder = os.path.join( 
        
               self.am_res_path, self.task_resource.res_dict['ckpt'][1]) 
        
           self.am_postnet = os.path.join( 
        
               self.am_res_path, self.task_resource.res_dict['ckpt'][2]) 
        
           # must have phones_dict in acoustic 
        
           self.phones_dict = os.path.join( 
        
               self.am_res_path, 
        
               self.task_resource.res_dict['phones_dict']) 
        
           self.am_stat = os.path.join( 
        
               self.am_res_path, 
        
               self.task_resource.res_dict['speech_stats'])

感觉保持原样比较统一，还是说一定要过了才能合并呢😂😂😂

zxcd · 2024-11-01T07:57:23Z

关于代码风格的问题，你可以使用pre-commit这个工具，使用参考：

pip install pre-commit
pre-commit run --file 你修改的代码

zxcd · 2024-11-06T06:24:28Z

LGTM. Thank you for your contribution!

luotao1 · 2024-11-06T12:22:35Z

hi, @SuiYunsy

非常感谢你对飞桨的贡献，我们正在运营一个PFCC组织，会通过定期分享技术知识与发布开发者主导任务的形式持续为飞桨做贡献，详情可见 https://github.com/luotao1 主页说明。
如果你对PFCC有兴趣，请发送邮件至 ext_paddle_oss@baidu.com，我们会邀请你加入~

Fix a bug of streaming_tts_server

4bb1970

Fix "Failed to get model related files" when using fastspeech2_csmsc_onnx model in streaming_tts_server.

mergify bot added the Server label Oct 25, 2024

Fix CodeStyle

4d430de

PaddlePaddle locked and limited conversation to collaborators Nov 6, 2024

PaddlePaddle unlocked this conversation Nov 6, 2024

zxcd merged commit 8279539 into PaddlePaddle:develop Nov 6, 2024
5 checks passed

paddle-bot bot added the contributor label Nov 13, 2024

GreatV mentioned this pull request Feb 28, 2025

PaddleSpeech 1.5.0 Release Note #3996

Closed

zxcd mentioned this pull request Jul 4, 2025

启动在线推理服务失败 #4094

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix a bug of streaming_tts_server #3865

Fix a bug of streaming_tts_server #3865

Uh oh!

SuiYunsy commented Oct 25, 2024

Uh oh!

CLAassistant commented Oct 25, 2024 •

edited

Loading

Uh oh!

zxcd commented Oct 30, 2024

Uh oh!

SuiYunsy commented Oct 30, 2024

Uh oh!

zxcd commented Oct 31, 2024

Uh oh!

SuiYunsy commented Oct 31, 2024

Uh oh!

zxcd commented Nov 1, 2024

Uh oh!

zxcd commented Nov 6, 2024

Uh oh!

Uh oh!

luotao1 commented Nov 6, 2024

Uh oh!

Uh oh!

Fix a bug of streaming_tts_server #3865

Fix a bug of streaming_tts_server #3865

Uh oh!

Conversation

SuiYunsy commented Oct 25, 2024

PR types

PR changes

Describe

Uh oh!

CLAassistant commented Oct 25, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

zxcd commented Oct 30, 2024

Uh oh!

SuiYunsy commented Oct 30, 2024

环境：

配置：

Uh oh!

zxcd commented Oct 31, 2024

Uh oh!

SuiYunsy commented Oct 31, 2024

Uh oh!

zxcd commented Nov 1, 2024

Uh oh!

zxcd commented Nov 6, 2024

Uh oh!

Uh oh!

luotao1 commented Nov 6, 2024

Uh oh!

Uh oh!

CLAassistant commented Oct 25, 2024 •

edited

Loading