Work at SourceForge, help us to make it a better place! We have an immediate need for a Support Technician in our San Francisco or Denver office.


EMC Arctic voices

Robert E
  • Robert E
    Robert E

    I am on a project for the University of Kansas and seriously need help with Espeak. Mbrola is very robotic for our text to speech needs, we found the Festival TTS and CMU Arctic voices work great yet the support for dynamic pitch, gap, and speed are not easily accessable…. We would like to know if Arctic can be support somehow for Espeak. We prefer Espeak…put the voice must sound natural or we are going to have to go festival somehow.

  • No, Festival and eSpeak are different text to speech engines.  The Arctic voices are part of the Festival speech engine.  There is no connection with the eSpeak project.

    If you want changes to the Arctic voices then you need to discuss in with Festival or Arctic developers.

  • Robert E
    Robert E

    Is there any way that you could help in naturalizing the voice, it a project for disabled kids. We need a voice that is very smooth, and easily understood.

  • Robert E
    Robert E

    We are using mb-us1 atm.

  • Robert E
    Robert E

    Here is what I have:

    name us-mbrola-1
    language en-us
    language en 3
    gender female

    phonemes en-us
    //dictrules 7 12

    stressLength 170 135 205 205 0 0  245 275

    pitch 80 200

    consonants 110 160

    flutter 7
    roughness 4
    formant 1 70 80 180 200
    formant 2 100 70 150 200
    formant 3 170 70 150 200
    formant 4 90 95 100
    formant 5 50 90 180
    formant 6 30 80 150
    formant 7 60 70 150
    formant 8 90 70 150
    stressAdd -100 -40 -80 -80 0 0 80 140
    formant 0 115 0 150
    voicing 70
    breath 75 75 50 40 15 10
    breathw 150 150 200 200 400 400

    mbrola us1 us_phtrans

  • If you want a voice which is "very smooth", then you need a different text to speech engine, not eSpeak.

    However, even though eSpeak voices sound artificial rather than human, I think the English voice is easy to understand.

    I prefer the standard "en" voice rather than mbrola such as "mb-us1".   I think it sounds clearer.

    It's possible that the the sound of mbrola voices could be improved if eSpeak used different lengths for the various mbrola vowel and consonant phonemes, but that would need some experimentation and program changes.

    Please note that attributes in a voice file such as "formant", "roughness", "flutter" and "breath" have no effect on mbrola voices.

  • Robert E
    Robert E

    Thanks for the tid bit :).