festvox / speech_tools Goto Github PK

I found the config.guess and config.sub files were updated in the source code, but some Linux distributions won't use the newest commit to build the package, only the release, so could you have a new release for speech_tools?

hello 👍

when i try to install i got this errorr

(make test)
wagon: -omp_nthreads ignored: not support in this build.
wagon script status: INCORRECT

and also i got

RMSE 37.2237 Correlation is 0.4856 Mean (abs) Error 28.7110 (23.7016)
wagon: -omp_nthreads ignored: not support in this build.

Would you consider to review a pull request that allowed the user to build speech_tools, festival and festvox using the meson build system (https://mesonbuild.com/)?

The current build system:

• has race conditions when building in parallel
• for festival and festvox it requires that speech tools is found at a specific parent directory
• does not honor --prefix
• does not allow out of source builds (i.e. the directory where speech tools is built must be the directory where the sources are)
• cross-compilation does not seem straightforward

meson is a nice (and modern) alternative. It should be possible to have both build systems (the current one and meson) and if meson works better eventually remove the current one in a future version.

Thanks for any feedback!

the repertory
speech_tools/wrappers/python seems missing in this version

I am finishing up some haskell bindings to the wonderful festival project. I include text to wave, and thus I am binding to the EST_Wave header as well. I see the formal int n of the method copy_sample specifies the row to copy from the underlying matrix type, but I don't know what that means for the sample. Is this just an index for channels in the EST_Wave? Thank you for your time.

This seems to currently have support for ALSA and PulseAudio on Linux (judging from #33), and the linked pull request seems to make life easier in adding support for other sound systems.

The Linux desktop is currently in the process of making a transition to PipeWire, which improves upon the previous APIs in the following ways:

• Capture and playback of audio and video with minimal latency.
• Real-time multimedia processing on audio and video.
• Multiprocess architecture to let applications share multimedia content.
• Seamless support for PulseAudio, JACK, ALSA, and GStreamer applications.
• Sandboxed applications support. See Flatpak for more info.

(https://pipewire.org/)

PipeWire maintains compatibility libraries for PulseAudio and ALSA, however these still have some of the limitations of their display server. Native PipeWire support would be ideal, particularly as I plan on utilizing this inside of a sandboxed environment, and PipeWire is the only multimedia API that handles that without punching a hole in the sandbox. Aside from that, the other benefits likely don't matter much, but still may be useful.

I don't know C/C++, so I probably can't be of much help, but I can try.

I'm running Linux Mint 19 Cinnamon 3.8.8

I'm trying to run main/na_record and I get the following error

~$ speech_tools/main/na_record -f 16000 -time 5 -o test.wav -otype riff

Linux: can't open /dev/dspfor reading

From searching online, it looks like OSS is no longer supported. I've tried installing some OSS compatibility modules, but still haven't had any success in running ./na_record

It seems like the /dev/dsp error has already been fixed in festival: https://bugs.launchpad.net/ubuntu/+source/festival/+bug/662630
I didn't have any issues with playback in festival either.

There's a few old workarounds listed for festival here: https://wiki.archlinux.org/index.php/Festival#Usage_with_a_Sound_Server

Is there any way I can apply those workarounds to festvox? Or are there any other methods to get recording/playback in festvox to work?

Hello,

From https://www.cstr.ed.ac.uk/projects/festival/download.html, it looks like the last release was in 2014.

Would it be possible for speech_tools to create tags for previous releases (or at least new ones going forward) in this repo, but also consider making a new release so distributions can easily pick up the fixes in master?

Thanks!

I would like to compile with python wrappers and the changes I made (simply by uncommenting) to config/config.in are:

## Uncomment following to enable building of wrappers
INCLUDE_MODULES += WRAPPERS

## Only set this if you *DO* want to run swig (for example to modify
## the wrappers yourself).  If so, the safest bet is to use the same
## version of swig as speech tools developers (download from
## http://www.swig.org/ (SWIG-3.02 last tested))
##
CONFIG_SWIG_COMPILER = /usr/local/bin/swig

# Languages to generate wrappers for. Currently: PYTHON
# PERL5 is no longer supported
CONFIG_WRAPPER_LANGUAGES = PYTHON 

# Language specific includes should be set to correct site paths
CONFIG_PYTHON_INCLUDES= -I/usr/include/python2.7/

and then
./configure
make info
make

and ran into this problem:

Making in directory ./wrappers ...
Making in directory wrappers/interface ...
Making in directory wrappers/interface/python ...
Making new directory wrappers/python
mkdir: python: File exists
make[2]: *** No rule to make target `update'.  Stop.
make[1]: *** [python] Error 2
make: *** [wrappers] Error 2

I'm using speech_tools 2.5, Apple LLVM version 9.1.0 (clang-902.0.39.2)
Does anyone have any advice? Thanks in advance!

In order to compile the festival repo I had to add the following to speech_tools/include/EST_math.h:

#define finite(X) isfinite(X)

For some reason it didn't cause an issue when compiling the speech_tools repo...

Hello.
When checking the logs after a Fedora mass rebuild using gcc 10, i came acros a link failure in speech_tools. You needed to add -fnocommon to the gcc args to fix it.
The error was:

Link Shared Library estools
if [ -n "" ] ; then libs='     -lncurses -lasound  -ldl  -lm  -lstdc++
-fopenmp ' ; fi ;\
gcc -shared -o libestools.so.2.5.0.1 shared_space/*.o  $libs
/usr/bin/ld: shared_space/siodeditline.o:(.data.rel.local+0x8): multiple
definition of `editline_history_file';
shared_space/editline.o:(.bss+0x28): first defined here
collect2: error: ld returned 1 exit status

Building speech-tools with make -j bumps out with errors. Only single core build seems to work.

os:gentoo
gcc version:6.4

./configure
make SHARE=1

/* Linux (and presumably Hurd too as Linux is GNU libc based) */           
/* Sorry I haven't confirmed this cpp symbol yet              */           
#if defined(linux)                                                         
#define isnanf(X) __isnanf(X)                                              
#endif 

In the case of musl based distributions, such as alpine linux, this assumption is incorrect.

I recommend the following fix,

#if defined(linux) && defined(__GLIBC__)
#define isnanf(X) __isnanf(X)
#endif

#ifndef isnanf	
#define isnanf isnan	
#endif

Hello.
Would it be possible to include the license text in a release tarball and as a separate file in the repository? It would help the packaging effords very much.

Putting this here in case anyone also gets the same error.

During ./configure:

checking for tputs in -lncurses... no

and during make:

/usr/bin/ld: siod/editline.c:403: undefined reference to `tgetstr'

It seems libncurses-dev is needed, not just libncurses, so

sudo apt install libncurses-dev

There might also be another make error:

/usr/bin/ld: cannot find -lcurses: No such file or directory

I have libncurses6, so this is solved by:

sudo ln -s /usr/lib/x86_64-linux-gnu/libncurses.so.6 /usr/lib/x86_64-linux-gnu/libcurses.so

Hi,

I am wondering if there might exist an integer underflow error:

1. comm_samples can be an any integer:
   

   speech_tools/speech_class/EST_wave_io.cc

   Line 796 in e2dcb2a

   if (ts.fread(&comm_samples, sizeof(int), 1) != 1)

2. If length is zero, then data_length can be a negative integer:
   

   speech_tools/speech_class/EST_wave_io.cc

   Line 839 in e2dcb2a

   data_length = (comm_samples-offset)*comm_channels;

3. So num_samples can be also a negative integer:
   

   speech_tools/speech_class/EST_wave_io.cc

   Line 857 in e2dcb2a

   *num_samples = data_length/comm_channels;

4. Call to fread with the negative integer:
   

   speech_tools/speech_class/EST_wave_io.cc

   Line 845 in e2dcb2a

   if ((n=ts.fread(file_data,get_word_size(actual_sample_type),

5. memcpy with a negative number can be vulnerable:
   

   speech_tools/base_class/EST_Token.cc

   Line 407 in e2dcb2a

   memcpy(buff,&buffer[pos],items_read*size);

Thanks for your time.