Instead of using the date the voice file was created use the version of Rockbox
the voice file was created for. That way it is possible to see from the
installation log if the installed voice matches the current Rockbox version.
Change-Id: I9fd7d40d718febafbc20579f049b9b94cfa5d573
Some language files have the < and > characters the original string (like "<All
tracks>") has included in the voice strings. These characters cause the
generated wave file to not contain any samples with SAPI, causing an error when
Rockbox Utility tries to encode the file.
Since those characters don't make much sense in the voice string remove them
prior to speaking. Currently affected languages are basque, chinese-simp, and
russian, for which this fixes voicefile creation.
Change-Id: Iac7ccd7921ccfb5ddc33645239a1f8f7e639e5aa
Instead of skipping empty voice strings when trying to voice the list of
strings drop them immediately when building the list of strings.
Change-Id: Ib2627fb4b5061916cf04d67b6b78f457122ed2cd
Pass the threshold value for wavtrim to the TalkGenerator object instead of
using the default value. Makes wavtrim to be actually used.
Furthermore, check the result of the wavtrim call and respect its return value.
Instead of calling the lame executable use libmp3lame directly. As result, this
simplifies the prerequisites for creating voice clips for Archos devices to
putting the library in the system's search path (Windows: put libmp3lame.dll in
the search path or the same folder RockboxUtility.exe is located in. Linux:
install the library using your systems package manager) and configuration.
This creates a notable encoding speedup on Windows (around factor 6 on my test
setup) and a small speedup on Linux (around factor 1.2).
The implemenatation currently has the following limitations:
- Only enabled on Windows and Linux. On OS X installing the correct dylib is a
bit nontrivial, so using the old command line based method is still in use
for now.
- The encoder parameters are currently hardcoded to use the same values the
build system uses.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@31634 a1c6a512-1295-4272-9138-f99709370657
Voicefile generation now can correct strings for the TTS system similar
to what voice.pl does. The current implementation has some limitations:
- only implemented for voicefile creation.
- the corrections file is built in and can't get changed.
- string corrections can be disabled in the configuration dialog.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@30628 a1c6a512-1295-4272-9138-f99709370657
Instead of hard coding the query part of the genlang url store the full url as
template and replace parts as done for other urls as well.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@29669 a1c6a512-1295-4272-9138-f99709370657
Move class-less functions in utils.cpp into a new Utils class and make the old
functions static. This prevents clashes with system C functions. Rename some
functions to avoid macro problems (check() is a macro on OS X). Split out the
RockboxInfo class into a separate file.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@25441 a1c6a512-1295-4272-9138-f99709370657
Simplify RockboxInfo file handling. Remove Detect::installedVersion() and Detect::installedTarget(), as those became wrappers around RockboxInfo without functionality.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@22237 a1c6a512-1295-4272-9138-f99709370657
- introduce a talkgenerator to better share code between voice and talk generation.
- name intermediate .talkfiles after the md5sum of the text. (prevents naming problems).
- do not directly use the logger, instead emit signals.
- move talkfile and voicefile generation to base/
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@21524 a1c6a512-1295-4272-9138-f99709370657
