support for --phoneme-input for having pure phonemes as input

[contentnation]

Feature to have pure phonemes as input, instead of text.
Can be (ab)used to create language mixing.
Example:
echo -n -e "ðɪs ɪz ˈɪŋɡlɪʃ wɪð ɐ fɹˈɛntʃ ˈaksənt" | piper -p -m fr_FR-upmc-medium
"This is english with a french accent" spoken by a french voice.

[fullymiddleaged]

Nudge, does this work now?

[contentnation]

@fullymiddleaged it works for me, if you have any cases where it breaks, please report it so it can be fixed.

[fullymiddleaged]

@fullymiddleaged it works for me, if you have any cases where it breaks, please report it so it can be fixed.

Oh, okidoke, I'll try again!

[HF353]

Hello, I was searching all around and finally looks like I found it. Just please can you axplain exactly what I shoud to do and where to type to use french model for English. I mean I want to use french accent. And I'm interesting to use all foreign models for accents. Is it possible? Thank you in advance!

[contentnation]

As written above:
echo -n -e "ðɪs ɪz ˈɪŋɡlɪʃ wɪð ɐ fɹˈɛntʃ ˈaksənt" | piper -p -m fr_FR-upmc-medium
-p for phoneme input instead of plain text
-m for the voice model and the matching accents.
plus the usual extra options like -wav output.wav
In theory it should work with every language/voice combination as long as the voice has the given phonemes.
Western phonemes with asian voice might not work correctly or the other way around.
To create the phonemes, you can use
echo "This is english with a french accent" | espeak-ng --ipa -q -v en
--ipa to output phonemes
-q to not create audio
-v to select language (if not given, english is used)

[HF353]

As written above: echo -n -e "ðɪs ɪz ˈɪŋɡlɪʃ wɪð ɐ fɹˈɛntʃ ˈaksənt" | piper -p -m fr_FR-upmc-medium -p for phoneme input instead of plain text -m for the voice model and the matching accents. plus the usual extra options like -wav output.wav In theory it should work with every language/voice combination as long as the voice has the given phonemes. Western phonemes with asian voice might not work correctly or the other way around. To create the phonemes, you can use echo "This is english with a french accent" | espeak-ng --ipa -q -v en --ipa to output phonemes -q to not create audio -v to select language (if not given, english is used)

Thank you so much for quick repley. I tried already command with no luck, but wasn't aware of espeak... so I'm gonna try and dig more to not distrub you, but I will return tomorrow for help sorry and big thanks!
P.S. I'm new on that, just few days, but managed already process of learning of model. And I tried to change in .json from fr to en-us. It's work, but unfortunateley some pronounce of course not like that (five instead Faive)...
Anyway, thank you once again I'm gonna learn!

[HF353]

I used the following command:
echo -n -e "ðɪs ɪz ˈɪŋɡlɪʃ wɪð ɐ fɹˈɛntʃ ˈaksənt" | piper -p -m fr_FR-upmc-medium.onnx -f .\phonemic.wav
She just reading letters kind of... Do you know what I'm doing wrong?

[contentnation]

I just checked the command. It worked fine here.
Just a few hints/questions. If you are on the wrong branch, it should not run and complain that -p is unrecognized.
So if works, you used the right branch/installation.
the output filename of .phonemic.wav is a little odd, but it does not break the logic. Maybe you are playing back the wrong file from a previous run?
Here is the audio created with the above command (and converted to ogg to save space).
https://userdata.contentnation.net/a5970e0955da4472b5f84a8dbb740273/phonemic.ogg

[HF353]

Yes, I wish to listen example (it's not playing, please check the link), as maybe it's fine... Basically I have the same sound if I will change in .json from "fr" to en-us....

[HF353]

Ah, okey, I saved the link and able to play. Thanks

[HF353]

Yes, I confirm! Absolutely the same result! I edit json file and rename to en-us.... Ah, too strong accent for my project. So need to find speaker that will read a text and then I will train the voice, I think it's only one best solution. Thank you very much for help!