4

I see that there are now Nitech HTS voices 2.3.2 available for download. On Ubuntu 20.04 festival 2.5.0 gets installed. How can I install and use latest Nitech HTS voices for festival on Ubuntu 20.04?

Mitar
  • 1,842
  • 1
  • 21
  • 25
  • Your question gives the impression that you want to use pre-built voices in festival. Please clarify that you want to train your own voices using Nitech HTS then build them to be used in festival if that is what you want. So that future answers will address what you need. – Raffa Mar 22 '21 at 09:16
  • It is not clear what you are asking... perhaps you are missing one or two version numbers in your description. – sancho.s ReinstateMonicaCellio Mar 22 '21 at 13:18
  • I hope to use pre-build voices in festival. [Here is a post](https://ubuntuforums.org/showthread.php?t=751169) how to do it for older versions, and [another one](https://darkshadow.io/2020/08/01/speech-synthesis-on-linux.html). But I am not sure how to do it for the latest HTS voices. – Mitar Mar 22 '21 at 16:01
  • There seems to be a bit of misunderstanding on your side... The *Nitech HTS 2.3.2 * which was released in 2017 is a speech synthesis system/engine which uses deep learning and neural networks to train speech modules based on datasets like [these](http://tts.speech.cs.cmu.edu/awb/cmu_arctic/)... The resulting modules can then be built for use with different speech engines including festivals. It is **not a group of voices** that you can download and use with festival. – Raffa Mar 23 '21 at 14:16
  • So you can either use the pre-built voices available in [this link](http://www.festvox.org/packed/festival/2.5/voices/) from my answer below as well as the ones provided by the links in your comment above or you can go ahead and build your own. – Raffa Mar 23 '21 at 14:20

1 Answers1

0

Notice

There seems to be a bit of misunderstanding on your side... The Nitech HTS 2.3.2 which was released in 2017 is a speech synthesis system/engine which uses deep learning and neural networks to train speech modules based on datasets like these... The resulting modules can then be built for use with different speech engines including festivals. It is not a group of voices that you can download and use with festival.

Pre-built voice modules are available on the Internet. So you can either use the pre-built voices available in the link below as well as the ones provided by the links in your comment above or you can go ahead and build your own and you can start from here.Please notice thee following on that page:

These distibutions include Festival CLUNITS based voices. bdl, slt, jmk and awb HTS based voices are available from available from http://hts.ics.nitech.ac.jp/ using Nagoya Institute of Technology's HTS HMM-based Speech Synthesis System.

So the voices are already pre-built and available for festival 2.5.0 since the year 2017 on this link

Please also keep in mind that setting the right environment, training and building is a lengthy process that might take days and a lot of resources and effort.

Adding pre-built voices to Festival (Version 2.5.0:release December 2017)

Festival voices go under /usr/share/festival/voices/ and the voices you refer to are linked here. To add and use new pre-built voices, please follow the instructions below:

  • Cleate a new directory for the new voices like so:

    sudo mkdir /usr/share/festival/voices/my_voices

  • cd to your Downloads directory like so:

    cd ~/Downloads

  • Donload desired voice from here like so:

    wget http://www.festvox.org/packed/festival/2.5/voices/festvox_cmu_us_aew_cg.tar.gz

  • Extract the downloaded voice file like so:

    tar -xf festvox_cmu_us_aew_cg.tar.gz

  • Copy the new downloaded voice located under festival/lib/voices/us/ in the extracted file to festivals voices like so:

    sudo cp -r festival/lib/voices/us/cmu_us_aew_cg/ /usr/share/festival/voices/my_voices/

  • Start festival like so:

    festival

  • List the available voices like so:

    (voice.list)

  • The output will look like this:

    (cmu_us_aew_cg kal_diphone)

  • Select the newly added voice by adding voice_ prefix before its name like so:

    (voice_cmu_us_aew_cg)

  • Test the new voice like so:

    (SayText "This is how you add new voices to Festival.")
Raffa
  • 24,905
  • 3
  • 35
  • 79
  • So how does this answer how to install Nitech HTS voices 2.3.2? – Mitar Mar 22 '21 at 07:06
  • @Mitar I updated the answer based on your comment above: "*I hope to use pre-build voices in festival*" – Raffa Mar 23 '21 at 14:55
  • Thank you for detailed notice. But my understanding is that [this link](http://www.festvox.org/packed/festival/2.5/voices/) does not contain pre-build voices from Nitech. The link they link to is dead. The answer here shows how to use CMU voices not Nitech. – Mitar Mar 23 '21 at 18:44
  • My understanding is also [from this page](https://ubuntuforums.org/showthread.php?t=751169), where they say: " These voices are produced by the HTS working group hosted at the Nagoya Institute of Technology. They have produced excellent quality voices which take up very little disk space. In terms of quality and size, probably the best (non-commercial) English voices availible for Festival. See the voice demo page (the *_arctic_hts are the voices of interest). Highly recommended. The voices are available on their download page." – Mitar Mar 23 '21 at 18:44
  • @Mitar "The link they link to is dead."... That is true and the reason is obvious... It is a research and volunteers work not backed up by commercial entity. So it is what it is one time the webpage is up and running and other times when the resources become scarce, it is dead! They do state somewhere on the festvox.org website that some voices are built with Nitech, but no guarantee... and I agree `arctic` voices have superior quality. – Raffa Mar 23 '21 at 20:03
  • `_arctic_hts` to be precise. Sadly those are not available on festvox site. So it seems those voices are not available anymore? They are not available as part of training data for Nitech system? – Mitar Mar 23 '21 at 20:08
  • @Mitar Training data sets at http://tts.speech.cs.cmu.edu/awb/cmu_arctic/ should work with HTS. As I remember [CMU](https://www.cmu.edu/) university were the pioneers in this and most open source speech training solutions including HTS make use of their work one way or another. – Raffa Mar 23 '21 at 20:16
  • But there are no artic hts voices pretrained available? – Mitar Mar 25 '21 at 04:09
  • @Mitar Apart from the old ones available on the Internet here and there, unfortunately no. At least I haven't seen one that clearly states it is trained using the latest HTS engine 2. The ones available like the ones you linked to are apparently built with older version HTS engine including version 1 which is available on Ubuntu for quite a while as a package called `htsengine`. – Raffa Mar 25 '21 at 06:56