Converts text to phonemes for Piper.
When using eSpeak phonemes, requires an espeak-ng fork with espeak_TextToPhonemesWithTerminator function.
This function allows for Piper to preserve punctuation and detect sentence boundaries.
Pre-compiled releases are available for download.
The piper_phonemize program can be used from the command line:
lib/piper_phonemize -l en-us --espeak-data lib/espeak-ng-data/ <<EOF
This is a test.
This is another test!
EOF
{"phoneme_ids":[1,0,41,0,74,0,31,0,3,0,74,0,38,0,3,0,50,0,3,0,32,0,120,0,61,0,31,0,32,0,10,0,2],"phonemes":["ð","ɪ","s"," ","ɪ","z"," ","ɐ"," ","t","ˈ","ɛ","s","t","."],"processed_text":"This is a test.","text":"This is a test."}
{"phoneme_ids":[1,0,41,0,74,0,31,0,3,0,74,0,38,0,3,0,50,0,26,0,120,0,102,0,41,0,60,0,3,0,32,0,120,0,61,0,31,0,32,0,4,0,2],"phonemes":["ð","ɪ","s"," ","ɪ","z"," ","ɐ","n","ˈ","ʌ","ð","ɚ"," ","t","ˈ","ɛ","s","t","!"],"processed_text":"This is another test!","text":"This is another test!"}Arabic diacritization is supported through libtashkeel (model included):
echo 'مرحبا' | lib/piper_phonemize -l ar --espeak-data lib/espeak-ng-data/ --tashkeel_model etc/libtashkeel_model.ort
{"phoneme_ids":[1,0,25,0,120,0,14,0,30,0,43,0,14,0,15,0,121,0,14,0,26,0,2],"phonemes":["m","ˈ","a","r","ħ","a","b","ˌ","a","n"],"processed_text":"مَرْحَبًا","text":"مرحبا"}
See src/test.cpp for a C++ example using libpiper_phonemize.
The piper_phonemize Python package is built using pybind11.
See src/python_test.py for a Python example.
Use Docker:
docker buildx build . -t piper-phonemize --output 'type=local,dest=dist'Find library and Python wheels in dist/
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent