Tts: Language independent text processing with text-to-unicode conversion.

[erogol]

TTS might be language agnostic, it the given text is processed by unicode representation instead of characters. That would give a much flexible solution for different languages without changing code.

[BlueTaurus]

doing it too for urdu dataset...

[erogol]

@BlueTaurus it is in the backlog, I've not yet started, unfortunately. But great, you give a hand.

[kurianbenoy]

I was planning to create a TTS system for Malayalam @ergol. Will it be possible to create such a TTS system based on Neural network based like this repository?

[erogol]

I don't know if the language has special properties but should be possible with TTS.

[alex73]

"Language-independent TTS" - does it mean "IPA-based TTS" ? (International Phonetic Alphabet)

[erogol]

No that was supposed to be Unicode based. But IPA is already enabled.

[richardburleigh]

I'm curious how IPA works so well despite being converted to ASCII. Is there some magic I'm missing? I haven't checked but I imagine there would be some duplication of phonemes that unicode could solve.

[sachaarbonel]

sentencepiece might be a good candidate to solve this

[reuben]

IPA is not converted to ASCII, it's mapped to integer labels. This issue is talking about training a model that takes Unicode text directly, without phonemization. Example: https://arxiv.org/abs/1811.09021

[richardburleigh]

Thanks for the reply @reuben! I got the impression from this line in cleaners.py, which ends up running the text through unidecode. Either way, love the progress on this.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discourse page for further help. https://discourse.mozilla.org/c/tts