Add Czech language

[SkaceKamen]

Description

Due to a bug in TTS package, coqui-ai/TTS#4098 idiap/coqui-ai-TTS#236 it's not possible to use Czech language right now. I've created an issue in the package repo, but found out it's no longer being maintained. I've made an issue & PR in the TTS library - once that's merged this issue will be resolved.

The issue is caused by quite recent release of num2words package which is used by TTS, see the PR here:
savoirfairelinux/num2words#587

So somehow downgrading that package would help, but I'm not sure if that's possible or even desired. The version that breaks the Czech language is https://github.com/savoirfairelinux/num2words/releases/tag/v0.5.14

Steps to replicate

Try to use Czech language, get a crash.

Stacktrace

File "/usr/local/lib/python3.10/site-packages/TTS/api.py", line 366, in tts_to_file
    wav = self.tts(
  File "/usr/local/lib/python3.10/site-packages/TTS/api.py", line 312, in tts
    wav = self.synthesizer.tts(
  File "/usr/local/lib/python3.10/site-packages/TTS/utils/synthesizer.py", line 406, in tts
    outputs = self.tts_model.synthesize(
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/models/xtts.py", line 410, in synthesize
    return self.full_inference(text, speaker_wav, language, **settings)
  File "/usr/local/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/models/xtts.py", line 479, in full_inference
    return self.inference(
  File "/usr/local/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/models/xtts.py", line 525, in inference
    text_tokens = torch.IntTensor(self.tokenizer.encode(sent, lang=language)).unsqueeze(0).to(self.device)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/layers/xtts/tokenizer.py", line 666, in encode
    txt = self.preprocess_text(txt, lang)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/layers/xtts/tokenizer.py", line 652, in preprocess_text
    txt = multilingual_cleaners(txt, lang)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/layers/xtts/tokenizer.py", line 573, in multilingual_cleaners
    text = expand_numbers_multilingual(text, lang)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/layers/xtts/tokenizer.py", line 562, in expand_numbers_multilingual
    text = re.sub(_number_re, lambda m: _expand_number(m, lang), text)
  File "/usr/local/lib/python3.10/re.py", line 209, in sub
    return _compile(pattern, flags).sub(repl, string, count)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/layers/xtts/tokenizer.py", line 562, in <lambda>
    text = re.sub(_number_re, lambda m: _expand_number(m, lang), text)
  File "/usr/local/lib/python3.10/site-packages/TTS/tts/layers/xtts/tokenizer.py", line 542, in _expand_number
    return num2words(int(m.group(0)), lang=lang if lang != "cs" else "cz")
  File "/usr/local/lib/python3.10/site-packages/num2words/__init__.py", line 98, in num2words
    raise NotImplementedError()

[ROBERT-MCDOWELL]

are you using our last git or v2.0.0 ? I see you are using your python system. are you running eb2ab with docker?
try to uninstall num2words and downgrade with pip install num2words==0.5.13. btw if you know where the issue comes from on v0.5.14 of num2words so you can try to create a PR on their repo

[SkaceKamen]

Sorry for not specifying details, I was running the docker version - the athomasson2/ebook2audiobookxtts:huggingface image, so I guess I was running the 2.0?

This can be easily resolved by grepping out the lang=lang if lang != "cs" else "cz" from the library itself or as you noted by pinning different version.

The root issue actually isn't with the num2words - they fixed their bug by replacing cz with cs which is the correct parameter. The TTS library should be fixed, either by pinning older version or by specifying correct lang, but unfortunately it's not maintained anymore.

Maybe I can persuade num2words to still support cz as legacy value? AFAIK no language has that code anyway...

[ROBERT-MCDOWELL]

the right iso fo czech is "cs", "cse", "cze", so it's on coqui-tts to modify it to 'cs', we are working with an active fork and you can tell them here https://github.com/idiap/coqui-ai-TTS

[SkaceKamen]

Oh, thanks for that, I was looking at the wrong library. I've made an issue & PR in the maintained one:
idiap/coqui-ai-TTS#236
idiap/coqui-ai-TTS#237

[ROBERT-MCDOWELL]

meanwhile if you find another TTS engine with the same or better quality than coqui-TTS so we can implement it.