Which TTS is better for the Dutch language?

64

Great, thank you! I’m thinking we should keep some of the sentences with English works to get a bit of the foreign phoneme coverage. Do you think those English words are common enough?

This must have been a way of getting the phoneme pair /iː o/ (wie is /w iː/). The /iː/ is from something like “anal[y]se” too. I see maybe 10 examples in the lexicon where this sound comes at the end of a word (like zei), and no cases where the pair occurs in a single word.

Sure, the coverage file contains all of the missing pairs and example words. If the example words all look unusual or foreign, it probably means the pair can be safely ignored.

65

Yes, maybe I was a bit too strict in weeding out all English words. Words like “Halloween”, “liken”, “showen”, “shop” and “bye” are probably common enough to include.

1 Like
66

Hi guys, I uploaded my Dutch dataset here.

Enjoy!

5 Likes
67

Awesome, thank you very much! I am training a new model now with your data :slight_smile:

1 Like
68

Quick update, here are some samples from the latest model: https://drive.google.com/drive/folders/1TXDr_ZeRks3fXDQmlIChlwFcDNK5ki4X?usp=sharing

I’m using GlowTTS and the Multiband MelGAN vocoder. I had to re-train from scratch this morning because of a mistake, so the vocoder isn’t sounding so great yet.

I’ll update the examples after it’s had a change to train overnight :slight_smile:

69

I’m looking forward to this! :smile:
I’m already ordering some Pi Zero for wakewords :crossed_fingers: in the house

1 Like
70

Updated the samples again with the vocoder about 50% done: https://drive.google.com/drive/folders/1TXDr_ZeRks3fXDQmlIChlwFcDNK5ki4X

Sounds much better!

1 Like
71

Awesome! Even at 50% the voice sounds great! Can’t wait for 100% :wink:

Would it be possible to do this for French also? Can we do this ourselves (recordings+training)?

1 Like
72

Definitely :slight_smile: The first step will be adding support for French to gruut. I expect this to be done by next week or so.

The recordings can be done with anything as long as you have WAV files and transcripts. With a little work, gruut will help select a small set of sentences from a large corpora (books, Wikipedia, etc.) that are maximally useful.

Training just needs a CUDA-enabled GPU that’s supported by PyTorch >= 1.5. I’m using a GTX 1060 6GB.

73

Updated the samples again with the vocoder about 50% done: https://drive.google.com/drive/folders/1TXDr_ZeRks3fXDQmlIChlwFcDNK5ki4X

4 of them are OK, but in the sentence “moeder sneedt zeven scheve sneden brood”, brood is
mispronounced.

kind regards,
Hugo

1 Like
74

Something cool I realized I could do this morning: https://drive.google.com/file/d/1PQneOLgfIyHqACsNj4x3FT1_Ypnuy6w0/view?usp=sharing

This is the rdh Dutch voice speaking an (accented) English sentence! Because I use IPA for both English and Dutch phonemes, I created a small mapping file that approximates the 14 or so “missing” phonemes from Dutch. I had to guess on some, but it works as a proof of concept :slight_smile:

So this means we could re-use some of the voices for other languages, until we get native speakers in that language. To get it right, though, I will need help from people who speak both languages. For example, I don’t really know how Dutch folks pronounce the “th” sounds from “thing” and “the”.

75

I don’t really know how Dutch folks pronounce the “th” sounds from “thing” and “the”

These sounds don’t exist in Dutch

kind regards,
Hugo

76

I substituted them with “s” and “z” sounds as a guess :man_shrugging:

What’s neat is that you could approximate a very specific accent with this approach.

77

Dutch people starting to learn English often mispronounce the “th” of “thing” as a plain “t” and the “th” of “the” as a “d”.

kind regards,
Hugo
p.s. started trimming the phrases, still 1300 to go…

1 Like
78

Much appreciated :slight_smile: I’ll get your model in the queue as soon as I get the files!

79

The first Dutch voice for Larynx is ready!

Special thanks to @rdh for donating his voice :clap:

It’s easiest to check out with Docker:

$ docker run -it -p 5002:5002 \
     --device /dev/snd:/dev/snd \
     rhasspy/larynx:nl-rdh-1

Then visit http://localhost:5002 to test it out :slight_smile:

You can use this from Rhasspy or Home Assistant by changing the run command slightly:

$ docker run -it -p 59125:5002 rhasspy/larynx:nl-rdh-1

Now you can tell Rhasspy/HomeAssistant that you’re using a MaryTTS server and point it to http://localhost:59125 (usually the default).

2 Likes
80

Loving it so far! :smile: Enormous thanks to @synesthesiam & @rdh

Just a few remarks, I don’t know if these are typical remarks that need manual finetuning.

Written numbers don’t come out well, 6 uur 46 vs. zes uur zesenveertig.
Example: GetTime intent

Some small things like OK is not correctly pronounced. However if you input “okee” the pronunciation is just fine.

81

For things like numbers and dates you should probably preprocess your text with something like Lingua Franca to convert them to words that are pronounceable by the TTS.

82

There are some undocumented features I’m still experimenting with, but I agree that in general a separate library should be used. Some of the features that are in there but disabled for now:

  • Currency recognition
    • “$100.12” (sort of works now)
  • Number types
    • “1_ordinal” becomes “first” in English
    • “1902_year” becomes “nineteen oh two” in English
  • Alternative pronunciations
    • “read_1” and “read_2” are pronounced like “red” and “reed” respectively

I also have the ability to list abbreviations for a language that are automatically expanded. I’ve got a list for English, like mr -> mister, but I don’t know any for Dutch.

83

Happy to help and expand on those lists.
If I remeber correctly Mycroft has something like a collaborative system on their website.
(translate.mycroft.ai)

I suppose we could do something similar with just a github directory per language and documentation on what is needed for completing a language.

1 Like