Skip to content

Latest commit

 

History

History
97 lines (83 loc) · 5.89 KB

espeak.md

File metadata and controls

97 lines (83 loc) · 5.89 KB
Raw

eSpeak-ng submodule

The submodule contained in the espeak directory is a cross platform open source speech synthesizer.

Building

NVDA has a custom build of eSpeak because not all components are required.

Background

The main authority on build requirements should be include/espeak/Makefile.am. The *.vcxproj files in include/espeak/src/windows/ can also be considered, however these are not always kept up to date.

We don't use the auto make files or the visual studio files, we maintain our own method of building eSpeak. Modifications will need to be made in nvdaHelper/espeak

  • sconscript for the build process.
  • config.h to set the eSpeak-ng version that NVDA outputs to the log file.

Updating the version used by NVDA

  1. Start from a clean branch off of NVDA master
    1. Check out the latest NVDA origin/master and create a new branch.
    2. Do a git clean to ensure the working directory is clean.
  2. Ensure submodules are up to date
    1. Synchronize submodules with git submodule sync
    2. Update submodules with git submodule update --init --recursive
  3. Checkout the new eSpeak-ng revision to try.
    1. Change to the include/espeak/ directory
    2. Do git fetch to get the latest from the espeak-ng repo
    3. Do git checkout origin/master or whichever espeak-ng ref you wish.
  4. Look at changes in Makefile.am and update our build.
    1. Diff Makefile.am with the previously used commit of espeak.
    2. Changes to Dictionary compilation should be reflected in espeakDictionaryCompileList
    3. Some modules are intentionally excluded from the build. If unsure, err on the side of including it and raise it as a question when submitting a PR.
    4. Modify the nvdaHelper/espeak/config.h file as required.
  5. Update our record of the version number and build.
    1. Change back to the NVDA repo root
    2. Update the /DPACKAGE_VERSION in nvdaHelper/espeak/sconscript
    3. Update NVDA documentation and changelog with eSpeak version and commit.
    4. Build NVDA: scons source
      • Expected warnings from eSpeak compilation:
        • On the first build after changes, all languages may show this warning. Our build intentionally compiles using the phonemetable.
        espeak_compileDict_buildAction(["include\espeak\espeak-ng-data\uz_dict"], ["include\espeak\dictsource\uz_list", "include\espeak\dictsource\uz_rules"])
Can't read dictionary file: 'C:\Users\sean\projects\nvda\include\espeak/espeak-ng-data\uz_dict'
Using phonemetable: 'uz'
Compiling: 'C:\Users\sean\projects\nvda\include\espeak\dictsource/uz_list'
      121 entries
Compiling: 'C:\Users\sean\projects\nvda\include\espeak\dictsource/uz_rules'
      35 rules, 26 groups (0)
  6. Run NVDA (set eSpeak-ng as the synthesizer) and test.
  7. Ensure that the log file contains the new version number for eSpeak-NG

Troubleshooting Build failures

If python crashes while building, check the log. If the last thing is compiling some dictionary try excluding it. This can be done in nvdaHelper/espeak/sconscript. Remember to report this to the eSpeak-ng project.

If the build fails, take note of the error, compare the diff of the Makefile.am file and mirror any changes in our sconscript file.

Known issues

Due to problems with emoji support (causing crashes), emoji dictionary files are being excluded from the build, they are deleted prior to compiling the dictionaries in the nvdaHelper/espeak/sconscript file.

Manually testing eSpeak-ng

If you wish to test eSpeak-ng directly, perhaps to create steps to reproduce when raising an issue for the eSpeak-ng project, consider feeding SSML directly to the executable provided by the project.

The eSpeak docs are worth reading. They describe the various (eSpeak-ng) command line arguments (note also speak-ng command line), and instructions to build an .exe from a commit of eSpeak, locally on Windows. However, historically the Windows build for espeak-ng hasn't been well maintained, with periods of build failures. It is also different from the build approach within NVDA.

  1. Install an (x86) release: https://github.com/espeak-ng/espeak-ng/releases espeak.exe is in C:/Program Files (x86)/espeak-ng/ Adding it to your path temporarily will make the following easier.
  2. Write out some SSML that demonstrates the problem. 
    $ cat /c/work/test-espeak-ssml.txt
<prosody pitch="+100">I am now speaking at an extra high pitch.</prosody>
<break time="1s"/>
<prosody pitch="1">I am now speaking at the default pitch.</prosody>
  3. Feed the SSML into espeak-ng.exe 
    $ cat /c/work/test-espeak-ssml.txt | ./espeak-ng.exe --stdin -m