tihu server weird respond

[mohamad-hasan-sohan-ajini]

Hi

I try to build and run tihu docker. Creating and running docker image using current Dockerfile:

$ sudo docker build . $ sudo docker run -p 50051:50051ab1fb002c73f error loading g2p model: -1mbrola: mbrola: No such file or directory � mbrowrap error: mbrola exited with status 1 Server listening on 0.0.0.0:50051

After checking Dockerfile, I realize that make ready was missed so the g2p model was not downloaded. Add make ready to Dockerfile, will change the log to:

$ sudo docker build . $ sudo docker run -p 50051:50051 ab1fb002c73f error loading g2p model: -1mbrowrap: voice samplerate = 16000 mbrola started. mbrowrap: voice samplerate = 22050 mbrola started. Server listening on 0.0.0.0:50051

But in both cases server respond with something non sense!

���������@����@����� �‏���
��������?�
���������������

Please help me to run tihu server. @b00f regards

[b00f]

@hajix Thanks for using Tihu and sorry for the inconvenience.

You are right, make ready was missed from dockerfile. however this only didn't fix the issue and g2p model didn't load correctly. I could see it from your log as well : error loading g2p model: -1

I fixed it at here https://github.com/tihu-nlp/tihu/commit/c19434fea7c301c093ba8e8b2f56ca5d3547af1b. It just needed to add local bin path to the global PATH. Dockerfile now should load correctly, please update your local code.

Two things might help you using tihu: 1- g2p is not updated for long time and has some issue regarding to tensorflow unsuitability: look #40 2- The lexicon file contains only 500 words. The full lexicon is not published yet.

If you have any client app for working with grpc, I appreciate if you can publish it here.

[mohamad-hasan-sohan-ajini]

thanks for your response...

Pull latest changes and building docker resolve the g2p model errors. but still get the same response from the server!?

$ curl 127.0.0.1:50051 @@ �
?s

p.s. if there is no constraint to use tf for g2p model, I'll be glad to build a model with pytorch and train it on zaya

[b00f]

but still get the same response from the server!?

You can't call gRPPC server like that, You need to write a gRPC client to call tihu server. I have added new targets to compile protobuf for go and js. here: https://github.com/tihu-nlp/tihu/commit/cf82a37581efe637600d1053db1b6fdeeb2e4324

It's so easy to write a js or go client for tihu server. Go ahead. Unfortunately I haven't time to do it right now. I will appreciate if you can send a PR.

example of go client for gRPC: https://github.com/grpc/grpc-go/tree/master/examples example of node.js client: https://github.com/grpc/grpc/tree/master/examples/node

I'll be glad to build a model with pytorch and train it on zaya

I will appreciate any help. Just make sure the tags are similar with what we have in tihu. Look here: https://github.com/tihu-nlp/tihudict/blob/master/tihu.phones