Models are missing in rpm build

[M-Stenzel]

Hi dsnote folks,

first of all I would like to thank you for this great piece of software!

I built the latest version (4.8.3) via the make_rpm.sh script, which works perfectly on openSUSE 15.6 (software package requirements needed to be adjusted due to the fact that the package names differ slightly from the fedora package names).

O. K. No issues with installing the rpm package, no crashes, no "bad behaviour", but ...

When configuring speech-to-text and text-to-speech I found out that I cannot find any speech to text model covering the German language (French and English are provided).

Moreover I cannot find e. g. the CrisperWhisper model, which availability is discussed here...

[(https://github.com//issues/223)]

How can I add the missing modules?

Martin.

P. S. Moreover, in the GUI, the button "Translator" (next to "Notepad") is greyed out. No chance to use this function...
In the "Languages and Models" section "Translator" is greyed out as well...

Any ideas?

[gcomes]

When configuring speech-to-text and text-to-speech I found out that I cannot find any speech to text model covering the German language (French and English are provided).

Probably the rpm you build does not have all the feature enabled.
Did you try packman's package?

Moreover I cannot find e. g. the CrisperWhisper model, which availability is discussed here...

You need the python modules installed for that.

[M-Stenzel]

When configuring speech-to-text and text-to-speech I found out that I cannot find any speech to text model covering the German language (French and English are provided).

Probably the rpm you build does not have all the feature enabled. Did you try packman's package?

Thanks for telling me, and indeed things got better.
I should have read this before

Moreover I cannot find e. g. the CrisperWhisper model, which availability is discussed here...

You need the python modules installed for that.

O. K. it is installed now, worked.

[gcomes]

When configuring speech-to-text and text-to-speech I found out that I cannot find any speech to text model covering the German language (French and English are provided).

Probably the rpm you build does not have all the feature enabled. Did you try packman's package?

Thanks for telling me, and indeed things got better. I should have read this before

Did you miss https://github.com/mkiol/dsnote#how-to-install ?

[M-Stenzel]

When configuring speech-to-text and text-to-speech I found out that I cannot find any speech to text model covering the German language (French and English are provided).

Probably the rpm you build does not have all the feature enabled. Did you try packman's package?

Thanks for telling me, and indeed things got better. I should have read this before

Did you miss https://github.com/mkiol/dsnote#how-to-install ?

O. K. I did read through the HOWTO - INSTALL

Right now I deleted the config directory (after the switch from script build of rpm to packman rpm) to make sure there is no problem here. I made sure there are no other remnants from the script built rpm package.

Some things to consider:

There are no patches in the packman source rpm - obviously not needed any longer?

Building from the packman source rpm failed here due to errors when building the openfst (vosk) external package . Maybe this is something to look for in the future.

I installed libtorch from the official repos - both from 15.6 devel:languages:python:backports and from science:machinelearning without any error, however when running speechnote in the terminal I receive this error message:

[E] ... libs_availability:362 - py error: ImportError: /usr/lib64/libtorch_python.so: undefined symbol: _PyCode_SetExtra

Still I cannot install Vosk-Tuda-DE model, and the Translator is still greyed out.

Martin.

[gcomes]

O. K. I did read through the HOWTO - INSTALL

The install HOWTO tells you to install the packman's rpm for openSUSE. Not to rebuild from the source rpm.

Right now I deleted the config directory (after the switch from script build of rpm to packman rpm) to make sure there is no problem here. I made sure there are no other remnants from the script built rpm package.

Some things to consider:

There are no patches in the packman source rpm - obviously not needed any longer?

The source rpm contains all the sources necessary to build the final package.

Building from the packman source rpm failed here due to errors when building the openfst (vosk) external package . Maybe this is something to look for in the future.

Packman's rpm build should work if you use the 'build' command or 'obs' command. If you try to use 'rpmbuild'
or 'make_rpm.sh' I'm pretty sure it wont work.

I installed libtorch from the official repos - both from 15.6 devel:languages:python:backports and from science:machinelearning without any error, however when running speechnote in the terminal I receive this error message:
[E] ... libs_availability:362 - py error: ImportError: /usr/lib64/libtorch_python.so: undefined symbol: _PyCode_SetExtra

That's because there is a version mismatch between the python libraries provided by speechnote and the one you installed from obs. Again, all the correct working python sources are included in the source rpm.
Using other sources will give you problems.

Still I cannot install Vosk-Tuda-DE model, and the Translator is still greyed out.

Does your CPU support sse4.1? If not the translator is not available.
Also if the CPU does not support avx, many features will be disabled.

You should install packman's binary rpm. The two rpm have all the libraries that speechnote needs, and there will be no need to install extra libraries from obs repositories.
If something does not work with packman's rpm, then I can help figure out what the problem is.