Monthly archives: April, 2016

Moresampler 0.6.3 Release

(Jun. 7, 2016 - Moresampler 0.7.0 is released!)

This is a minor revision which fixes several crashes reported to occur on previous version (0.6.2).

Download Moresampler 0.6.3 here

For bug reporting, please go to Moresampler Bug Report Page.


Moresampler 0.6.2 Release

(Jun. 7, 2016 - Moresampler 0.7.0 is released!)

First of all, I'm constantly receiving reports saying Moresampler works bad with some certain voicebank and the voice sounds very, very harsh or noisy as if it's growling. Here is the problem explained:

It's simply because Moresampler failed to analyze the pitch of some audio files. The pitch analysis also includes the detection of voiced/unvoiced status, i.e. "whether it is vowel or consonant or silence at a particular time". When a voiced region is mis-detected as an unvoiced region, anything bad could happen. Unfortunately so far there doesn't exist an algorithm that achieves 100% detection rate. A way to get rid of this is to manually edit .mrq files, as shown in the tutorial. Anyway, it's still better if error rate could be reduced so you don't need to spend hours editing .mrq by hand.

So what I've been doing in the past 6 months is to constantly make changes to Moresampler every time I receive such a bug report. However once I fix an error on one voicebank, new pitch errors show up on some other voicebanks, pretty much like playing whack-a-mole and I see people wondering why "new Moresampler gets worse than old version".

Now I've made up my mind to take a much more systematic approach: evaluating Moresampler's performance on 360 samples randomly chosen from 24 voicebanks. Then I inspected the samples causing pitch detection errors one by one, improved the algorithm and ran the test again (and again and again...). Currently there are still some errors, but the error rate becomes at least ten times lower. This means that In most cases Moresampler will run perfectly, though occasionally you may need to tweak it a little bit.


The Complete Moresampler Tutorial

Since the very initial release, numerous features have been integrated into Moresampler, but the old tutorial somehow still targets at the now-deprecated 0.2.0 version. To help you take full advantage of Moresampler, here I've written a much more comprehensive tutorial, or a reference, if you'd like to call it that.

The current version of this tutorial is written for Moresampler 0.7.1. Some features may not work on previous versions.

Contents
  1. Overview
  2. Setting up and basic usage
  3. Configuration file
  4. Flags
  5. Tips and troubleshooting
  6. Tips for voicebank authors

Moresampler 0.6.1 Release

(Apr. 18, 2016 - Moresampler 0.6.3 is released!)

This version is a revision which fixes two bugs in Moresampler 0.6.0. Besides, it will update .llsm files to the latest version so you don't need to delete them by hand.

.mrq format has been updated. In the new version, each data entry has a timestamp keeping track of last modified time and a status number indicating whether or not the entry has been manually modified.

Moresampler 0.6.1 has a new configuration: auto-update-llsm-mrq. When turned on, it checks the last modified time of .wav and corresponding .llsm file. If .wav has been modified, Moresampler will automatically update .llsm and .mrq files. This could be a helpful feature for voicebank developers.

Recently I'm working on a comprehensive tutorial for Moresampler and it is finally released along with Moresampler 0.6.1. Check it out here.


Back from Redbird Hacks 2016

This afternoon I just came back from Redbird Hacks, a 36-hour hackathon at Illinois State University, which is actually the first hackathon I've been to (though a month ago I visited HackIllinois without actually participating in).

My team was formed shortly after the event began, consisting of four U of I freshman and sophomore students. We spent a great deal of time discussing what to build with ideas ranging from a multiplayer version of Pong on Pebble Time to a Donald Trump chat bot with speech recognition & synthesis. Finally we settled down on an Android app that reads a paper tape using camera and plays music notes according to dots drawn on the tape. This is quite similar to a music box, except being a digital version based on image recognition. Not bad for a team of freshman and sophomores, we won the third prize: a $40 loud speaker for each member.