Can't install nemo

[858514558]

Building wheel for pynini (setup.py) … error
error: subprocess-exited-with-error
fst::script::ShortestPath(fst::script::ShortestPathArgs1*)’
41 | void ShortestPath(ShortestPathArgs1 *args) {
| ^~~~~~~~~~~~
/usr/include/fst/script/shortest-path.h:41:6: note: template argument deduction/substitution failed:
extensions/_pywrapfst.cpp:58426:89: note: candidate expects 1 argument, 3 provided
‘FST_FLAGS_fst_error_fatal’ was not declared in this scope; did you mean ‘FLAGS_fst_error_fatal’?
76506 | FST_FLAGS_fst_error_fatal = 0;
| ^~~~~~~~~~~~~~~~~~~~~~~~~
| FLAGS_fst_error_fatal
error: command ‘/usr/bin/aarch64-linux-gnu-gcc’ failed with exit code 1
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed building wheel for pynini

[858514558]

full.log (136.2 KB)
详细日志请查看附件

[dusty_nv]

Hi @858514558, Nemo isn’t formally supported on Jetson and there may be some patching required to get it to build. YMMV, but you can see some Dockerfiles and pre-built container images I have for it here:

[858514558]

ths. i will try it

[858514558]

I would like to inquire about where to find a Mandarin Chinese Riva TTS model

[dusty_nv]

Hi @858514558, you might want to post on the Riva forum for more help with TTS Zh model - here is what I could find:

• TTS Zh Fastpitch HifiGan SFSpeech | NVIDIA NGC
• How do I train a Chinese TTS model? · NVIDIA/NeMo · Discussion #4473 · GitHub

[858514558]

I am trying to use llamaspeak (https://github.com/dusty-nv/jetson-containers/tree/master/packages/llm/llamaspeak). I want to use Riva for Chinese ASR (Automatic Speech Recognition) and TTS (Text-to-Speech) functionalities. After changing the language configuration, ASR is working fine, but there is no corresponding Chinese model for TTS. I found a tutorial that suggests using “nemo2riva” to convert a Mandarin model into a Riva model. So I started installing Nemo and encountered the aforementioned problem. In fact, my original requirement was simply to find a Chinese model that is compatible with Riva.

[dusty_nv]

If my nemo-for-JetPack container doesn’t work with it, then converting the models you could probably best attempt on an x86 machine somewhere using NGC nemo or riva container, and then download the converted model to your Jetson.

Also, the LLM you use would also need to be trained in Mandarin, or otherwise using En<->Zh translation.