I’m new to TTS and highly appreciate the great work of you all! Thanks!
I’m trying to work into the topic and have therefore installed the self-contained package that runs an HTTP API from the Released Models page in a docker container. I am able to synthesize speech, however, some phrases result in the crash of the application. My feeling is that longer phrases are more likely to throw an exception. For example:
“They were carried along a dark tunnel, which seemed to be taking them right underneath the castle, until they reached a kind of underground harbour, where they clambered out on to wet rocks and slippery pebbles, soaked by a weak of rain.”
The error on the web frontend is given as:
Error: NetworkError when attempting to fetch resource.
The docker container logs say:
[INFO] 172.17.0.1 - - [28/Apr/2020 06:28:46] "GET / HTTP/1.1" 200 - python: /pytorch/third_party/ideep/mkl-dnn/src/common/scratchpad.cpp:63: mkldnn::impl::global_scratchpad_t::global_scratchpad_t(size_t): Assertion scratchpad_ != nullptr’ failed.`
A quick google search brought up this pytorch issue. Is it safe to assume, that the error is not directly related to TTS but a general issue with pytorch?
For the sake of completeness my used dockerfile is shown here:
FROM python:3.6 RUN apt-get update && \ apt-get -y install espeak libsndfile1 wget RUN wget https://github.com/reuben/TTS/releases/download/ljspeech-fwd-attn-pwgan/TTS-0.0.1+92aea2a-py3-none-any.whl RUN pip install TTS-0.0.1+92aea2a-py3-none-any.whl ENTRYPOINT [ "python", "-m" ] CMD [ "TTS.server.server" ]