Replies: 1 comment
-
Looking at sample.txt, I see it contains: "[[ˈoʊˌsətɛi̯χ]] [[ˈoʊˌŋəkʊ]] [[ˈlikʏχkə]] [[ˈpœy̯ɣi]] [[ˈaːpɑmyɦ]] [[ˈpʊkəŋœy̯]]" -- IPA text enclosed in double square brackets. After running it through lexconvert with the output set to "espeak" I get: ***@***.***:]] ***@***.***] [['li:kk@]] [['pi:]] [['A:pA:m]] ***@***.***]". If I change the output option to x-sampa, I get "oU"s@%tEi oU"N@%kU li"kk@ pi" a:"pAm ***@***.***".
espeak-ng uses Kirshenbaum. That's what you've produced here. We also have an SSML implementation available with espeak-ng -m, but it's buggy and unrealiable.
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I am working on a text-to-speech related application that produces input in a phonetic representation. In my current implementation, I am using Amazon Polly. But I would like eSpeak-ng as the default (no-cost) option.
At this moment, I have (temporarily) given up on direct integration due to issues with getting my Windows .NET (Visual Studio 2022) project to properly reference the eSpeak-ng libraries after successfully building them. So, I will call the eSpeak-ng directly, passing in command line arguments.
I am looking for documentation, or at least informal help, on how to submit a string of IPA, X-SAMPA, Conlang X-SAMPA, or if needed another format.
FWIW, I did look back into the ancient (well several months old) history on my Linux box and found that I had conducted the following experiment:
cat sample.txt | lexconvert --phones2phones unicode-ipa espeak | espeak-ng -g 10 -w sample.wav
Looking at sample.txt, I see it contains: "[[ˈoʊˌsətɛi̯χ]] [[ˈoʊˌŋəkʊ]] [[ˈlikʏχkə]] [[ˈpœy̯ɣi]] [[ˈaːpɑmyɦ]] [[ˈpʊkəŋœy̯]]" -- IPA text enclosed in double square brackets. After running it through lexconvert with the output set to "espeak" I get: "[['oU,s@tEi:]] [['oU,N@kU]] [['li:kk@]] [['pi:]] [['A:pA:m]] [['pUk@N]]". If I change the output option to x-sampa, I get "oU"s@%tEi oU"N@%kU li"kk@ pi" a:"pAm pU"k@N".
(FWIW: I cannot use lexconvert because it does not accurately or completely translate IPA to X-SAMPA due to its primary writer's use of British English (I'm guessing Received Pronunciation) which lacks a full set of phonemes, notably vowels, and merges several. I will handle the conversion out of the IPA I use internally into whatever phonetic representation I need to feed espeak-ng in my code.)
I would appreciate any assistance with:
FYI: When I work with Amazon Polly, I utilize the SSML "phoneme" tag, but I note that this is not currently implemented in espeak-ng. Even though this project may end up including either a fork of espeak-ng, or work that gets submitted back to the project, I am expecting that work to be towards a limited-use "universal" voice that can support a broad set of phonemes beyond what any currently supported voice can support. I am unlikely to also have the time to implement support for the "phoneme" tag.
Ron Oakes (RonBOakes)
ron@ron-oakes.us
ro421@mynova.nsu.edu
Beta Was this translation helpful? Give feedback.
All reactions