Problem with italian dataset creation

[diegobernagozzi]

Hi everyone,

im trying to create an italian dataset for melotts italian training, but when i start the training it produces only unintelligible sounds. I think that the issue it could be in the files like metadata.list.cleaned, train.list and val.list.

After running

python3 preprocess_text.py --metadata data/example/metadata.list

the phonemes of many phrases in the file are like this:

../../audio_1/Untitled_MIC_1_995.wav|Italian|IT|Ogni giorno mi avvicino sempre di più al mio sogno|_ ˈ o ɲ ɲ i ˈ d ͡ ʒ o r n o m i a v v i t ͡ ʃ i n o ˈ s ɛ m p r e ˈ d i p j u a l ˈ m i o s o ɲ o _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 5 8 2 5 5 7 3 3 2 4 4 1

gruut generate this symbol ͡ that the training code treat as a stand alone symbol, and maybe cause problem, because if i running this gruut test code (from gruut repo):

from gruut import sentences

ssml_text = """<?xml version="1.0" encoding="ISO-8859-1"?>
<speak version="1.1" xmlns="http://www.w3.org/2001/10/synthesis"
    xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xsi:schemaLocation="http://www.w3.org/2001/10/synthesis
                http://www.w3.org/TR/speech-synthesis11/synthesis.xsd"
    xml:lang="en-US">
<s>Today at 4pm, 2/1/2000.</s>
<s xml:lang="it">Ogni giorno mi avvicino sempre di più al mio sogno</s>
</speak>"""

for sent in sentences(ssml_text, ssml=True):
    for word in sent:
        if word.phonemes:
            print(sent.idx, word.lang, word.text, *word.phonemes)

the italian output is like this:

1 it Ogni ˈo ɲ ɲ i
1 it giorno ˈd͡ʒ o r n o
1 it mi m i
1 it avvicino a v v i t͡ʃ i n o
1 it sempre ˈs ɛ m p r e
1 it di ˈd i
1 it più ˈp j u
1 it al a l
1 it mio ˈm i o
1 it sogno s o ɲ o

it seems that the ͡ is not a stand alone symbol, and i'd like to know how to handle this symbol in the melo tts code.

Thanks in advance,

Diego