Training a TTS model with a language that doesn't have a supported phoneme_language

Archives TTS (Text-to-Speech)
#1

I’d like to train a TTS model in an Indigenous language that uses an orthography that doesn’t have a supported phone set. Would I be able to change the training data from the language’s orthography into it’s IPA representation and train it that way? Example below:

wavFile1|həloʊ wɜːld
waveFile2|siː spɑːt ɹʌn
waveFile3|tədeɪ ɪz ɐ naɪs deɪ

If that’s the case would I set the follow values in the config.json file?

    "use_phonemes": false,           // use phonemes instead of raw characters. It is suggested for better pronounciation.
"phoneme_language": "en-us",

Any help is appreciated!

#2

if you have IPA representation directly in the dataset, then just use character based training but give IPA characters as the char list (not phoneme char list).

#3

Ok thanks for the reply so would my config.json look like this?

    "characters":{
    "pad": "_",
    "eos": "~",
    "bos": "^",
    "characters": "aɑæɔtʃɛeɪdɡʔhidʒknoʊustwjz: ",
    "punctuations":" ",
    "phonemes":""
},

Also would the settings below be correct?

// PHONEMES
"phoneme_cache_path": "mozilla_us_phonemes_2_1",  // phoneme computation is slow, therefore, it caches results in the given folder.
"use_phonemes": false,           // use phonemes instead of raw characters. It is suggested for better pronounciation.
"phoneme_language": "en-us",     // depending on your target language, pick one from  https://github.com/bootphon/phonemizer#languages

Thanks again for the help!

#4

yes if these are the only phoneme chars for russian