Uberduck

and even sans that feeding raw kanji to tacotron is a Bad Idea

even google i think used utf-8 bytes instead

but translit cleaners deals well enough with kana

I already have the code in the notebook to convert the transcription files to romaji

Oh and here's the transcript with the romaji code

speakers/0000_jvs001/parallel100/wav24kHz16bit/VOICEACTRESS100_001.wav|mata, Dong Si noyouni, Wu Da Ming Wang toHu bareru, Zhu Yao naMing Wang noZhong Yang niPei sarerukotomoDuo i. |0

Is there a python script that converts kanji to kana I could run before using that command then?

thats chinese. it converted the kanji to chinese.

I noticed that, that's an interesting technicality

pyopenjtalk can convert everything to katakana

So if I wanted to write this it would probably be

```path = "/content/project/transcription.txt"

!pip install pyopenjtalk
from pyopenjtalk import kana

with open(path) as f:
  text = f.read()
with open(path, "w") as f:
  f.write(unidecode(text))```

you'd have to split the transcript this time

```py
path = "/content/project/transcription.txt"

!pip install pyopenjtalk
from pyopenjtalk import g2p

with open(path) as f:
  text = f.read().splitlines()
text = [i.split("|") for i in text]
text = [[i[0], g2p(i[1, kana = True]), i[2]] for i in text]
with open(path, "w") as f:
  f.write("\n".join(["|".join(i) for i in text]))
```

I got an error for invalid syntax for g2p(i[1, kana = True]), specifically the equal sign

its `g2p(i[1], kana = True)` not `g2p(i[1, kana = True])`