Aeneas – tweaking the character table

  • LorenH
    Participant
    Post count: 27
    #64056 |

    I had not messed with tweaking the character substitution table in Aeneas before, but learned from one of our linguists that it’s possible to get major improvement in results with some tweaks. Today I tried having it ignore the glottal stop and the underline on vowels. I mapped x to s for the language I was working on. I went from not very good verse level timing to quite good phrase level. Maybe some sharing on what works would be helpful. The Zapotec linguist had found that the pre-nasalized consonants (nd, etc) needed to be mapped to the basic consonant with out the nasal on the front. Any other bits of advice out there? I’m sure this is very language specific, but certain kinds of phonetic issues and orthographic issues may be common.

  • Paul Federwitz
    Participant
    Post count: 13
    #69857 |

    Would you be willing to write up a bit more about what you did? I think it would be very helpful to others.

  • LorenH
    Participant
    Post count: 27
    #75021 |

    I am still experimenting with this, so don’t have definitive answers. Some things I am currently trying:

    map the glottal stop character to nothing
    map x to s
    map j to h
    map g to h
    map gu to g
    map ts to s, or maybe t
    mb to b
    ng to g
    nd to d
    tl to t
    barred i to u instead of i

    This is all going to be very language specific according to how the orthography works in that language. I do seem to be getting somewhat better results on problem languages, but not perfect. Maybe with more experience, I can get this going better.

  • LorenH
    Participant
    Post count: 27
    #87459 |

    I have a language that I tried tweaking the table on, but I just can get it to synch right. Does anyone have ideas on how to improve the Aeneas results on hard to do languages?

  • jheath
    Participant
    Post count: 5
    #87533 |

    Once you have Aeneas installed, from a command prompt you should be able to type:

    espeak “phrase to say”

    There are various command line parameters, like -x shows you the “phonetics” of what it is saying. You can specify the voice to use with -v (should correspond to the voice you use when running Aenaes – I think esperanto would be -ves). More details on the parameters are here:

    http://www.mit.edu/afs.new/athena/system/i386_deb50/os-ubuntu-9.04/usr/athena/share/doc/espeak/commands.html

    I found that I could play various phrases and figure out some changes for Aenaes which made the text sound more natural, like I changed a long “ee” to a simple “e”, because it sounded more like the language.

  • jheath
    Participant
    Post count: 5
    #87534 |

    Sorry, the code for Esperanto is “eo” (you can see the code when you select the language as you are running Aeneas in SAB). So if you are using Esperanto as your language in Aeneas, to hear what Aeneas is producing as speech (and then comparing to your audio to perform text synchronization), type this at a command prompt:

    espeak -x -veo “Allah maragaani min beet abuui da”

    In my case, I think I get slightly better-sounding results if I get rid of the doubled vowels:

    espeak -x -veo “Allah maragani min bet abui da”

    So those are some of the replacements that I make when running Aeneas.

    So if you want to find some replacements that help your synchronization, I would recommend playing around with espeak on the command line (with the voice you use in Aeneas) to see what regular changes to your text can make the voice output sound a little more natural. The good news is that you don’t actually have to make it sound particularly natural, but I think the closer you can get to the “cadence” of natural speech, the better the synchronization will be.

    • LorenH
      Participant
      Post count: 27
      #87976 |

      Thanks for the good tips. I had played just a little with eSpeak, but this gives me some better ideas of how to go about it.

You must be logged in to reply to this topic.