When not compiling a voice with the system but using the load-functionality (e.g. using cst_cg_load_voice), the global list mimic_voice_list does not work intuitively.

Voice loading can be seen in mimic_main.c:

    if (mimic_voice_list == NULL)
        mimic_set_voice_list(voicedir);
    if (desired_voice == 0)
        desired_voice = mimic_voice_select(NULL);

mimic_voice_select searches the voice list and if it is not found, it tries to load the voice from file - but not if the list is NULL/empty in the first place. In that case mimic_voice_select returns NULL.
Therefore in the example tool mimic_main.c, the global mimic_voice_list is directly accessed.

A possible workaround to load the voice is to ignore the lists and use:

   mimic_init();
   mimic_add_lang("eng",usenglish_init,cmu_lex_init);
   ...
   voice = mimic_voice_load(path);

Probably mimic_voice_select could be modified so that it also works with empty lists too.
Also, there are no functions to remove a voice from a list and delete it.

oss_audio: failed to open audio device /dev/dsp

This is on a stock Ubuntu machine.

Some words are pronounced incorrectly.

The two that come to mind in my testing are
Atheism and Penis

Atheism is pronounced by Mimic

A thigh ism

Now Theism, Theist and Atheist are pronounced correctly so this is a bit puzzling why atheism is pronounced differently.

Penis is pronounced by mimic like the words

pen is

it should be pronounced like

pee nis

Now I tested this out with a few voices and had identical results.

Hi,

One of the most common complaints about mimic is the long compile times it requires. The main cause for that is that the mycroft voice is compiled embedded in the mimic binary, instead of being loaded on runtime from a file.

We don't load the voice from a file on runtime because it is too slow. However, if we were able to improve the voice loading functions then we could stop embedding it at compilation time. So far, @forslund has made some improvements in #85 but still there is room for improvement.

I need someone to test a command that loads the mycroft voice from file. Then that person needs to compile mimic with a patch that may improve voice loading performance slightly and then check if there is a significant improvement or not.

1. Download and compile the development version

We will disable all the embedded voices (with --disable-voices-all) to make compilation much faster:

git clone https://github.com/MycroftAI/mimic.git
cd mimic
git checkout development
./configure --disable-voices-all
make

2. Test the timings (copy the output of this command)

time ./mimic -voice voices/mycroft_voice_4.0.flitevox  -t "" a.wav

3. Clean up:

make distclean

4. Try the patched version:

git remote add zeehio https://github.com/zeehio/mimic.git
git checkout zeehio/cg_maybe_faster_load
./configure --disable-voices-all
make

5. Test the patched version (copy the output of this command)

time ./mimic -voice voices/mycroft_voice_4.0.flitevox  -t "" a.wav

Thanks to anyone who can help on this

Problem installing from git on Fedora 25, build_host_setup_fedora.sh completes without errors. ./dev_setup.sh fails when compiling mimic

checking for gcc... gcc
checking whether the C compiler works... no
configure: error: in `/home/james/mycroft-core/mimic':
configure: error: C compiler cannot create executables

config.log says error 77

-----------

confdefs.h.

-----------

/* confdefs.h */
#define PACKAGE_NAME "mimic"
#define PACKAGE_TARNAME "mimic"
#define PACKAGE_VERSION "1.2.0.1"
#define PACKAGE_STRING "mimic 1.2.0.1"
#define PACKAGE_BUGREPORT "https://github.com/MycroftAI/mimic/issues"
#define PACKAGE_URL ""
#define PACKAGE "mimic"
#define VERSION "1.2.0.1"

configure: exit 77

Any help would be appreciated :-)

First and foremost I think it is important to have solid documentation around this project. I encourage those contributing to try and put as much as possible in the Github wiki in this repo.

Some things I'd like to have in the Wiki:

1. How to easily install.

2. How to use the software.

3. How to adjust things like speed, cadence, and tone.

I think these are good questions to get us started, once we develop an understanding of the software and document it. I think we can move on from there.

Some people, institutions and licenses define source code as the preferred format for modification.

In mimic we have some data that is shipped as C code. We distribute this C code, but it is auto generated and it is not the preferred source for modification. As an example, the lexicon (#15), the letter to sound rules or the voice models.

Far from being a simple licensing issue, this impacts the repository size and more importantly our ability to understand where the code comes from and how to fix issues fast. On the other hand having data preprocessed in the repository reduces the build dependencies (festival speech synthesis would be required) and the build time (voice models are already created).

For a fast build time and quick testing with ability to fix bugs we could include the cmulex in source format AND its autogenerated equivalent C source code (both the lexicon and letter to sound rules created from it). But this would increase the repository size a lot if a correction is made (we already are at about 500 MB!).

Another option could be to keep all raw data under version control using git-lfs (Large file support) [1] but I don't have experience in that so I don't know how easy it is to set up.

Ideas and personal preferences are very welcome

[1] https://git-lfs.github.com/

It seems as if the required mingw32 and ming32-runtime packages are not available for 16.04. I installed mingw-w64 (no separate run-time) but this fails and I don't know what to try next.

FYI - native compilers and runs ok. I assume that I can use a single file tree for both native and cross builds.

Thanks for any help.

Introduction (to be expanded)

• waveform synthesis techniques: there are many ways to synthesize a waveform: diphone (e.g. cmu_us_kal16 in mimic) cluster units, clustergen (eg cmu_us_slt, vid_us_ap), HTS... Each technique has pros and cons.
• hts voices.
  • hts Advantages: small footprint good quality.
  • hts issues: a bit more robotic than unit selection based engines, this can be improved with a better vocoder. Not easy to find free vocoder implementations.
  • There is a version of Flite with hts engine support here
• HTS support would be good for mimic due to the small voice footprint.
• Finding and integrating a good vocoder in mimic could make a difference in voice quality with small footprint.
• Some discussion was going on here #61.

(I will copy a summary of the discussion here)

If you want to work on this just say it. If no one steps up I will do it whenever I have time (I would rather focus on multilingual support).

Mimic should have an audio test. I suggest using a pure tone so we have a "unit test". If creating the wave is an issue for you, you can synthesize text and adjust timings accordingly.

• Print on the screen the compiled audio output module (alsa or pulseaudio or portaudio...)
• Test that there is sound in the output
  • Create a cst_wave made of a pure frequency (sine at 440Hz during 1 second)
  • Play the cst_wave
• Test that it takes at least 40 seconds to run (audio not interrupted before the end)
  • Create a cst_wave made of the concatenation of:
    • Sine at 440 Hz during 1 second
    • Sine at 880 Hz during 1 second
    • Sine at 1760 Hz during 36 second
    • Sine at 880 Hz during 1 second
    • Sine at 440 Hz during 1 second
  • Play the cst_wave, measure the time it takes to be played.
• Test audio_write does not block: Assert that there is no audible pause between the two waves:
  • Wave 1:
    • Sine at 440 Hz during 3 second
    • Sine at 880 Hz during 3 second
    • Sine at 1760 Hz during 6 second
  • Play wave 1
  • sleep 0.2 seconds (simulates that Wave 2 generation takes time, but less than Wave1 seconds. This sleep runs while audio is still playing)
  • Wave 2:
    • Sine at 880 Hz during 3 second
    • Sine at 440 Hz during 3 second
  • Play wave 2
• Test that audio can be interrupted. (Measure that time running < 12 seconds)
  • Create a cst_wave made of the concatenation of:
    • Sine at 440 Hz during 4 second
    • Sine at 880 Hz during 4 second
    • Sine at 1760 Hz during 4 second
  • Play the cst_wave, measure the time it takes to be played.
  • sleep 2 seconds
  • Interrupt audio

The best way to present this software is to have a live demo on the website where you can write text, select a voice and click "speak".

See this example: http://tts.speech.cs.cmu.edu:8083/

Is the work finished? Do you need any help Where can I find information?

I would like to add to mimic continuous integration infrastructure.

I would like on every commit to:

• Know if all the tests pass
• Know the percentage of code covered by tests
• Have a static code analysis report looking for potential issues
• Have a report of other checks (code style...)

To achieve that I plan to:

• Use Travis CI (To build the project)
• Use coverity for static code analysis
• Use code climate for the rest

I have some experience with these tools but before implementing this I would like to know:

• If any of you know of better or other tools
• If MycroftAI as an organization is happy to have that setup
• Your opinion, comments and feedback

make -k test fails with this output:

Makefile:79: warning: overriding recipe for target 'multi_thread'
Makefile:76: warning: ignoring old recipe for target 'multi_thread'
make[1]: *** No rule to make target 'test'.
Makefile:130: recipe for target 'test' failed
make: *** [test] Error 2

In command line I tried running ./mimic -t where string was about 1800 characters long, but Mimic didn't read the whole thing. It got to about one third of the text and then just stopped mid-word at about 460 character.
All characters were either letters, numbers or punctuation, combined into proper grammatical sentences.
I tried with other long texts and got same result.

I use updated Fedora 23 x86_64 and BASH as command line interpreter.

Can you provide an Android version via F-Droid? Seems there are no good FOSS TTS systems for Android.

This is not really an issue with mimic but probably still worth considering:
I found that the function names in src/audio/audio.c frequently clash with other libraries (e.g. Acapela TTS). Perhaps prefixing is an option here?

Also, it seems that when defining CST_AUDIO_NONE (and e.g. using another audio framework), only audio_stream_chunk in https://github.com/MycroftAI/mimic/blob/master/src/audio/au_streaming.c requires audio.c - and it seems to be an example function only.
So probably surrounding audio_stream_chunk with #ifndef CST_AUDIO_NONE would make sense? (or even move that function to an example file).

From synaptic, the package manager:

libmimic is an open source video decoding library for decoding Mimic V2.x-
encoded content (fourCC: ML20), which is the encoding used by MSN Messenger
for webcam conversations.

The codecs metapackage mint-meta-codecs on Linux Mint installs it, as a dependency of some of the gstreamer multimedia plugins.

I suggest renaming mimic libraries from libmimic to libttsmimic, any other ideas?

(I was going crazy until I found out I was not linking to the right library!)

The engine works on Mac OS X but cannot generate sound output (only file output is possible). Many users would be grateful if also the Mac could speak.

pcre2 has a smaller scope (just regular expressions) and supports UTF-8. It is easier and faster to build than ICU and as it is a C library we would not need a C++ compiler, for instance on cross-compilation.

Not only that, but also given its size we can consider embedding a copy of the pcre library in mimic and, in case the library is not found, compile it.

pcre is used by nginx, julia and R among other projects, so it is quite popular

This is on my to-do list after I finish the thesis, feel free to work on it if you want

I see the following test failures.

Test project...                 [ FAILED ]
  lex_test_main.c:69: Check strcmp(val_string(val_car(syl)), tok) == 0... failed
  lex_test_main.c:69: Check strcmp(val_string(val_car(syl)), tok) == 0... failed
  lex_test_main.c:69: Check strcmp(val_string(val_car(syl)), tok) == 0... failed
Test atypical...                [ FAILED ]
  lex_test_main.c:69: Check strcmp(val_string(val_car(syl)), tok) == 0... failed

Git revision: 0513650
Configuration: ./configure --prefix=/usr --with-audio=alsa
Test command: make -k test
Host system: x86_64 Arch Linux

I am in the build process (make) and it seems to take too long. It is stuck on a line that reads:
CC lang/vid_gb_ap/libttsmimic_lang_vid_gb_ap_la-vid_gb_ap_cg_01_mcep_trees.lo

What is wrong? Is it an error? Should I wait?
I am running it on GalliumOS on an Acer Chromebook 14.

speech-dispatcher is a high-level device independent layer for TTS and is the universal interface that is used by GNOME and KDE for desktop speech integration.

https://devel.freebsoft.org/speechd

It supports a plugin interface so I believe a mimic speechd plugin would be a good initial integration for Mycroft/minic integration for general desktops.

Develop support for the French language.

A section should be added to the README.md detailing the development model. Should at least include

• Always create pull requests against the Development branch
• Coding style requirements

In addition commit log message style might be nice but may be pushing the bureaucracy one step too far =)

I'm not sure what is going on here becasue some machines have no problem with compiling, while others do. It seems to be in part due to heavy resource usage.
I will return with error snippets as soon as I can.

It have passed some hours from the ./dev_setup.sh from 'mycroft-core' and when it reaches the part where works in 'mimic' i got tons of these warnings. In my raspberry pi 3 which is a single board computer, and with 4GB Swap, it finished compiling or something like that in about 2 hours or maybe less. And now in my AMD64 Mini Barebone PC (Giada A51B: AMD Dual-Core T56N, 4GB RAM, 128GB-SSD) is taking so much more and now i only see that warnings:

Warning: initialization discards 'const' qualifier from pointer target type [-Wdiscarded-qualifiers]

My RPi3 has Raspbian with PIXEL (the lastest and updated)
and my other PC has Linux Mint 18, updated with apt-get upgrade, etc...

Is that normal?
Thanks

After setting up a virtual machine without alsa or portaudio OSS is detected as audio despite that no OSS support actually exist. And when running mimic the error message "oss_audio: failed to open audio device /dev/dsp is shown.

On modern linux boxes OSS can still be used, there is an alsa compatibility layer and Arch linux has a guide for installing OSS as a native driver. If we intend to keep the OSS support (for linux) some more advanced detection than

AC_CHECK_HEADER(sys/soundcard.h,
              [AUDIODRIVER="oss"
               AUDIODEFS=-DCST_AUDIO_LINUX])

is needed.

One possibility is to check if /dev/dsp actually exists. It is created by the alsa-oss compatibility layer and is the hard coded sound deviced used by mimic's OSS layer. It does feel a bit dirty...

When building mimic on a system without libicu the configuration step finishes without issues but the make step fails when trying to build objects requiring libicu

Need to develop the ability to support the Spanish language.

When starting mimic with -l for looping, ctrl-c causes the Shutdown requested! message to be infinitely printed without quitting.

Current version works pretty damn good, and if we kick off a stable release (it seems stable right now). We can use that version for Mycroft.

Support for the German language.

I've tried quite a few times to create voices with Mimic on my rPi3.

I'm able to run ./bin/mimic -t "Hello. Doctor. Name. Continue. Yesterday. Tomorrow." -o hello.wav
but when i run aplay hello.wav I hear no audio, but any other wav file plays just fine.

I installed this with mycroft-core using ./dev_setup.sh

I started doing some work on cleaning up the README, but we really need an installation and usage blurb in there that isn't overwhelming.

PASS: unittests/hrg_test
PASS: unittests/regex_test
FAIL: unittests/string_test
PASS: unittests/token_test
PASS: unittests/voice_select
PASS: unittests/wave_test
PASS: unittests/lex_test
PASS: unittests/lts_test
PASS: unittests/nums_test
============================================================================
Testsuite summary for mimic 1.2.0.2
============================================================================
# TOTAL: 9
# PASS:  8
# SKIP:  0
# XFAIL: 0
# FAIL:  1
# XPASS: 0
# ERROR: 0
============================================================================
See ./test-suite.log
Please report to https://github.com/MycroftAI/mimic/issues
============================================================================
make[3]: *** [Makefile:4225: test-suite.log] Error 1
make[3]: Leaving directory '/tmp/yaourt-tmp-stardiviner/aur-mimic/src/mimic-1.2.0.2'
make[2]: *** [Makefile:4333: check-TESTS] Error 2
make[2]: Leaving directory '/tmp/yaourt-tmp-stardiviner/aur-mimic/src/mimic-1.2.0.2'
make[1]: *** [Makefile:4618: check-am] Error 2
make[1]: Leaving directory '/tmp/yaourt-tmp-stardiviner/aur-mimic/src/mimic-1.2.0.2'
make: *** [Makefile:4110: check-recursive] Error 1
make: Target 'check' not remade because of errors.
==> ERROR: A failure occurred in check().
    Aborting...
==> ERROR: Makepkg was unable to build mimic.

Seems one testing is not passed.

Hi all,

Mimic though alsa currently forces the audio output to be a single channel. I believe this device doesn't have support for single channel audio (yeah...odd). I get the following error:

$ ./bin/mimic -t "hello"
audio_open_alsa: failed to set number of channels to 1. Invalid argument.

In verbose mode doesn't seem to give any further information.

This is some information on the Jabra speaker according to aplay (happy to provide any other information):

$ aplay -L
null
    Discard all samples (playback) or generate zero samples (capture)
pulse
    PulseAudio Sound Server
sysdefault:CARD=USB
    Jabra SPEAK 510 USB, USB Audio
    Default Audio Device
front:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    Front speakers
surround21:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    2.1 Surround output to Front and Subwoofer speakers
surround40:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    4.0 Surround output to Front and Rear speakers
surround41:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    4.1 Surround output to Front, Rear and Subwoofer speakers
surround50:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    5.0 Surround output to Front, Center and Rear speakers
surround51:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    5.1 Surround output to Front, Center, Rear and Subwoofer speakers
surround71:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    7.1 Surround output to Front, Center, Side, Rear and Woofer speakers
iec958:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    IEC958 (S/PDIF) Digital Audio Output
dmix:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    Direct sample mixing device
dsnoop:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    Direct sample snooping device
hw:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    Direct hardware device without any conversions
plughw:CARD=USB,DEV=0
    Jabra SPEAK 510 USB, USB Audio
    Hardware device with all software conversions

$ aplay -l
**** List of PLAYBACK Hardware Devices ****
card 1: USB [Jabra SPEAK 510 USB], device 0: USB Audio [USB Audio]
  Subdevices: 1/1
  Subdevice #0: subdevice #0

Github allows to set the default branch where pull requests should go, I believe we should set it to "development" instead of "master", if we want new PR to go to "development" by default.

The preference can be changed in https://github.com/MycroftAI/mimic/settings/branches

Clustergen (CG) "flitevox" files are binary dumps created by cst_cg_dump_voice.
This restricts compatibility with voices trained on other systems.
Endianness is checked in cst_cg_read_header but just throws an error instead of byte-swapping.
There is a number of uses of e.g. sizeof(int), so different word sizes matter.

To reduce startup time the voice loading should be optimized.

Possibilities

• Reduce number of calls to cst_fread()
• Increase read buffer using setvbuf()
• Restructure flitevox-file to allow for more efficient reading
• inline commonly used functions (?)

Currently the engine works on Linux, Android, Mac OS X, and Windows. If we can make it work on iOS in relatively short order that would be a big win.

Is there a way for it to mimic a person's voice by using that person's voice samples? Thank you.

Getting home from vacation trying to build the mimic development branch I got the following error

CDPATH="${ZSH_VERSION+.}:" && cd .. && /bin/bash /home/ake/projects/c/mimic/config/missing aclocal-1.14 -I m4
/home/ake/projects/c/mimic/config/missing: line 81: aclocal-1.14: command not found
WARNING: 'aclocal-1.14' is missing on your system.
         You should only need it if you modified 'acinclude.m4' or
         'configure.ac' or m4 files included by 'configure.ac'.
         The 'aclocal' program is part of the GNU Automake package:
         <http://www.gnu.org/software/automake>
         It also requires GNU Autoconf, GNU m4 and Perl in order to run:
         <http://www.gnu.org/software/autoconf>
         <http://www.gnu.org/software/m4/>
         <http://www.perl.org/>
Makefile:1752: recipe for target '../aclocal.m4' failed
make: *** [../aclocal.m4] Error 127

This is because aclocal is version 1.15 on my ubuntu system and not 1.14. On my debian VM I get the same issue with a fresh checkout of the project unless I install automake.

The Makefile seem to include rules for building the aclocal.m4 and the dependencies may be considered updated when checked out and this trigger the error in question.

I see two solutions:
1: remove the files generated by autogen.sh and the procedure for building shall include running ./autogen.sh (I've prepared a branch for this named autogen)

2: The rules should exclude aclocal.m4 (This I haven't quite figured out how to accomplish)

It seems that current sample rate is 8khz,
Can I modify this to 16khz ?
Thanks.

Git commit cd1a36b added uuid (https://pypi.python.org/pypi/uuid/1.30) as a dependency that doesn't support python3 and is unmaintained (last release was 2007). The python standard library already includes a rfc4122 generator so that should be used instead (https://docs.python.org/2/library/uuid.html).

Maybe in flitevox form? I wasn't able to find any.

Hi! I was trying to install mycroft-core on my ubuntu 17.04 desktop (64-bit), and an error prompted saying:

/usr/bin/ld: warning: libicui18n.so.58, needed by ./.libs/libttsmimic.so, not found (try using -rpath or -rpath-link)
/usr/bin/ld: warning: libicuuc.so.58, needed by ./.libs/libttsmimic.so, not found (try using -rpath or -rpath-link)
/usr/bin/ld: warning: libicudata.so.58, needed by ./.libs/libttsmimic.so, not found (try using -rpath or -rpath-link)

I couldn't find the mentioned package on the package manager.
Is there anything I can do to make this work!
Thx!