Vietnamese phonemes

[sultanovazamat]

Hi, thanks for the great library!

I am trying to run inference using vietnamese models, but this is what I get for both of the available models:

echo 'Trăm năm trong cõi người ta, chữ tài chữ mệnh khéo là ghét nhau.' | scripts/piper --model voices/vi/vi-25hours-single-low.onnx --output_file welcome.wav Traceback (most recent call last): File "/usr/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/usr/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/workspace/piper/piper/src/python_run/piper/__main__.py", line 74, in <module> main() File "/workspace/piper/piper/src/python_run/piper/__main__.py", line 63, in main wav_bytes = synthesize(text) File "/workspace/piper/piper/src/python_run/piper/__init__.py", line 72, in synthesize phoneme_ids.extend(self.config.phoneme_id_map[phoneme]) KeyError: '1'

Do you have any ideas why is this happening? Thanks.

[trunglebka]

Vietnamese here, Piper's Vietnamese voice is not good. It sounds like a foreign saying Vietnamese. I think the problem is poor dataset and phoneme set does not fit Vietnamese

[synesthesiam]

@sultanovazamat A fix is coming soon for this! As @trunglebka said there are problems with the Vietnamese phonemes in Piper. I don't know how good espeak-ng's phonemes are (where Piper gets them), but there was a bug that was discarding some phonemes during training -- specifically the tones that Vietnamese would rely on.