Jerry You comments

Results 18 comments of


                                            Jerry You

When GRPC protobuf generates the pb.go file, does it necessarily define the message structure as a pointer? Is there any way to avoid this

a pointer or not matters a lot?

When GRPC protobuf generates the pb.go file, does it necessarily define the message structure as a pointer? Is there any way to avoid this

i know every time a pointer is null. but i don't understand that nested structures cannot be modified at present. Can you give a example of this?

wiki数据，括号中的内容丢失

> 也不是都丢失了，好像括号中包含了英文的话，就会丢失，能否修复下这个问题呢？为何不自己使用的时候处理一下，找到问题然后提pr

DropOut PlaceHolder should be a scalar, how to construct it with tensorProto

> all the request parameters as list or `tensor` that's quite right. I am afraid causing that error is because of using a wrong model. But with a list like...

Using beam as a library, or at least allow imporing its packages

i am still confused about how to deploy my own cluster beams, no time to think about the project structure about it. Anyone can tell me how to deploy prod...

prod deployment docs

ok, I see. It's not easy, It's important to be familiar with using beam in local mode firstly , then to think about the cluster if it's necessary. And It...

Collecting metrics

i got this problems: `io.netty.handler.codec.http2.Http2Exception: HTTP/2 client preface string missing or corrupt. Hex dump for received bytes: 16030100850100008103031a44d4a944f84807b840c9d654`

> unimrcp检测到activity是连续的0了，就自动关闭了channel，不会识别后5秒有话语的语音。但我发现unimrcp的声音检测效果不好，而且敏感度的调节参数（好像是0-256）跳变的比较厉害我也发现了这个问题，修改默认的阈值能有一定改善，但是达不到满意的效果 > 因此我自己实现的时候自己编译了一个C语言写的VAD模型（使用GMM训练），使用动态库调用，加了一些后处理，加强了说话结束的识别效果，结果令人满意。这个有没有开源，参考参考

你好，我想请教一下你是如何实现用户不说话是结束识别的？

已经弃用了unimrcp，改用freeswitch加载定制的语音识别模块so文件，这样只需根据第三方语音识别api做处理，很大程度方面修改。