Monthly archives: February, 2016

Moresampler 0.3.1 Release

This revised version finally pushes the "multi-process conflict problem" to an end, which means Moresampler can finally run in parallel! If you own a dual-core machine (I believe your machine has at least 2 cores, as of 2016) and you have a UTAU plugin such as presamp which allows multiple instances of resampler.exe to run in parallel, you get a 2x boost for both analysis and synthesis! Further, there's no limit on the maximum number of processes running at the same time (I've tested this feature on a 16-core server and it worked perfectly well).

Another nice thing is the new option supported in moreconfig.txt - "analysis-anti-distortion on/off". This option overcomes the physical limit faced by many voicebanks where background noise or insufficient volume level interferes the actual signal, and resulting in "gross" or "coarse" sounding speech after pitch shifting (thanks Aline for pointing this out!). When turned on, Moresampler will automatically correct the unstable speech parameters. Here's an example of audio output without and with "analysis-anti-distortion" turned on.


Moresampler Bug Report Page

This blog post is dedicated to collect bug reports for Moresampler, in fact, not only bug reports but also general feedback on any kind of dissatisfaction about Moresampler (e.g. "it sounds dim/sharp/coarse/muffled/trembling/robotic/noisy/whispering/popping etc. on ... voicebank"). You are free to leave comments below and I'll be notified once a new comment appears here.

Usually fixing a bug/tweaking takes less than or around a week and will result in a new revision (e.g. 0.3.0 -> 0.3.1). Larger amendments or addition of features with some implementation difficulties may take longer time and will be scheduled to release in the next minor (e.g. 0.3.1 -> 0.4.0) or major (e.g. 0.3.1 -> 1.0.0) version (seriously, only very, very large changes will result as a new major release).


How to run UTAU on Linux

Obviously we need wine to run Windows programs on Linux, and indeed many people have managed to start UTAU on their distros. But here instead of reiterating how to setup wine, I'd like to present you the way to solve the encoding problems that hinder you from loading Japanese voicebanks.

Preliminaries

First you need to understand what exactly are the "encoding problems" we are facing. UTAU is a Windows software specifically designed for Japanese systems. The software itself along with associated files (e.g. voicebanks, configurations, readme...) uses SHIFT-JIS encoding. On Windows, you need to specify the default encoding for non-unicode programs to be SHIFT-JIS to get the program working, otherwise you get a bunch of mojibakes. This also applies for Linux, which uses UTF-8 by default.

When you extract a Japanese voicebank from a .zip file in some non-SHIFT-JIS environment, you also get a bunch of mojibakes, for some similar reason. But .rar files would be fine because RAR encodes file names in Unicode.


Moresampler 0.3.0 Release

After one month of intense research and development, here comes Moresampler 0.3.0.

Download Moresampler 0.3.0 here.

Although version 0.3.0 doesn’t differ much from the previous 0.2.3 version in terms of number of features added, the synthesis engine has undergone substantial upgrade and around 50% of the codes has been rewritten.

The new version comes with a new file format – the “desc.mrq” file, which is similar to .frq frequency tables used by most UTAU resamplers. The purpose of this file is to facilitate manual editing of pitch curves (though the editor for .mrq file is not ready yet). But the role of .mrq is slightly different in that .frq files are loaded on each time the resampler starts while .mrq file is loaded only once, which is when Moresampler analyzes the wave and generates its .llsm file. In fact, .mrq works as a “pre-analysis result” so as to help Moresampler correctly analyze the wave, getting rid of pitch and voicing detection errors. Once .llsm is generated, .mrq is no longer used. Even though, it’s better not to remove .mrq at this stage, since they are still needed when Moresampler 0.4 is out some time in the future.


Previous releases of Moresampler

0.2.3 (Jan. 4, 2016) Download

  • Bug fix: crashes caused by memory overflow when frequency upperbound is greater than sampling rate.
  • Bug fix: incorrect interpolation at the boundary between voiced and unvoiced frames.

0.2.2 (Jan. 1, 2016) Download

  • Bug fix: clips produced under resampler mode when output sampling rate is greater than analysis sampling rate.

0.2.1 (Dec. 15, 2015) Download

  • Bug fix: synthesis mode (under wavtool mode) not triggered for usts with lots of trailing R notes.
  • Bug fix: disproportional gain after pitch shifting.
  • Bug fix: a minor bug in wav output.
  • Bug fix: maximum noise frequency may be different among llsm files.
  • Improved feature: higher time resolution for sinusoidal analysis.

Porting my website onto WordPress framework

The major motivation behind this slightly tedious work is that I would be able to write some release notes on a blog-like website for future releases of Moresampler, instead of awkwardly leaving a few links unexplained. Secondly, I can release myself from the burden of writing html codes by hand, wasting my life on aligning the spaces before <li>, <p>, and <br>. Thirdly, visitors would be able to leave comments right on the page. Finally, this paragraph now has accomplished its mission of being a test post, and it shall be put under the “uncategorized” category.