voice synthesis features

[xoxos]

yes, but if you want them, you're going to have to sort through them and determine which ones don't have english parallels that list doesn't help me because i can imagine saying 'jour' a dozen different ways. even if the wav samples played, one voice is not enough for familiarity with a phoneme.

i don't mean to be rude, but if you want french phonemes, you will have to have a french and english speaker make the determination as to which phonemes are unique. i have looked at the lists and i am simply not qualified to do it.

recording another set of phonemes (say adding 20 new vowels) is not going to happen if you can find say 5 amendments, i can do that.

thanks birrbits

[spacedad]

is there a hard north-english 'u',a bit like a german 'u'?

[stanlea]

I can't do that before 10 days, sorry.

[antoine.bertier]

Using additional resources such as http://www.ipasource.com/extras/diction ... Charts.pdf and quickly comparing with http://en.wikipedia.org/wiki/IPA_chart_for_English I came with the following list of phonemes needed in french and not present in english:

Consonant:
- r prononced phonetically [r] as in Paris
- gn prononced phonetically [ɲ] as in montagne

Vowels:
- u prononced phonetically [y] as in tu
- é prononced phonetically [e] as in ses or blé
- eu prononced phonetically [ø] as in peu
- ain/in prononced phonetically [ɛ̃] as in demain
- en/an prononced phonetically [ɑ̃] as in campagne
- on prononced phonetically [õ] as in mon
- ui prononced phonetically [ɥ] as in suis
- un prononnced phonetically [œ̃] as in un, brun

And here is the IPA phonetic chart with an mp3 for the prononciation of each symbol: http://web.uvic.ca/ling/resources/ipa/c ... IPAlab.htm

All french speakers are welcome to comment or complete.

[xoxos]

as said, for correct integration with the set, i should say them, and even though i have heard french spoken, i'd probably want to hear say 4 different people pronouncing them to get the correct idea. i know i'm not confident with my french 'r'.

[stanlea]

as said, for correct integration with the set, i should say them, and even though i have heard french spoken, i'd probably want to hear say 4 different people pronouncing them to get the correct idea. i know i'm not confident with my french 'r'.

that makes sense - if you can wait a little, I can provide different recordings of that stuff.

[xoxos]

i can always amend the vst later.

tbh after hearing french in the contest entries, i think for best results it would be a better idea for someone to record their own formant set, and take a day off to do the analysis sometime. that way it would be done correctly.

[stanlea]

i can always amend the vst later.

tbh after hearing french in the contest entries, i think for best results it would be a better idea for someone to record their own formant set, and take a day off to do the analysis sometime. that way it would be done correctly.

if you can have several versions, it's good - for the analysis, how to make it ? i'm not sure I'm able to do it

[antoine.bertier]

i can always amend the vst later.

tbh after hearing french in the contest entries, i think for best results it would be a better idea for someone to record their own formant set, and take a day off to do the analysis sometime. that way it would be done correctly.

if you can have several versions, it's good - for the analysis, how to make it ? i'm not sure I'm able to do it

Indeed how do you do the analysis ? And I would need to find a microphone first if I want to record anything Or could I just use the internal mic from my wife's macbook ?

[xoxos]

the following steps are necessary for my process

the first step is to clearly identify the phoneme set for implementation, and distinguish between phonemes with a single formant, and those that change, for instance, to my ears, the french 'r' in 'paris' sounds almost like it always starts more like a soft 'w' and transitions to an 'r', meaning it might be better to handle it as two sets of formants.

for instance most of the klaat voice synthesizers use a single formant for the long 'a' as in 'day,' whereas it clearly changes to more of an 'e' sound towards the end, which is why i've implemented it as two formants.

i order my data:
single formant vowels
consonants
weird consonants (like 'zh' and 'kh')
double formant vowels ordered a1 a2 b1 b2 c1 c2..


step 2: the entire phoneme set must be recorded. i used my laptop's soundcard, as long as there's no background noise and it's clear.

i used a comfortable midrange pitch for my vowels (around 220Hz - i did analyse at higher and lower pitches, and decided to go with the midrange analyses) and said each one a few times to get a good average.

for consonants, i'd try and isolate them, and also record them at the end and beginning of words. by analysing your files you can see the ms you blend between phonemes, which is useful info when using the vst. ultimately the formants for consonants are a bit arbitrary because they're too short and complex to analyse as such.

step 3: praat is easy to install and has clear instructions on opening sound files for formant analysis. it's cumbersome because you have to press different function keys for each piece of data and close the windows for each one (f1 does formant 1 frequency, f2 formant 2, et c.)

you probably want to get comfortable with looking at your vowels in praat and seeing the variation and averaging. for analysis, you want to do the following:

make a brief .wav for each phoneme (short files quicker to load in praat for fft analysis)

for each vowel formant (eg. no fricatives, just smooth vowels with no noise) isolate and loop one wavecycle of your voice until it sounds 'perfect' for the vowel. loop it several times and save it as the wavefile.

step 4: there are three major classes of info i use (there are other minor things too )
formant frequency for each of 4 bands
amplitude of each formant
bandwidth for each formant

praat does bandwidth analysis, but it takes too long to press all those buttons

i opened the wav files in wavosaur and did an fft analysis of the selection, identified the frequency bands and recorded their amplitudes from 0 to 1 on a linear scale.

the q settings are values from 0 to 1 which are scaled the same as synthedit's sv filter (or any sv filter) (except synthedit is 0 to 10 instead of 0 to 1) and these can only really be determined by familiarity with this scale, so i'll probably have to do that.

most of the vowel q values are .6 to ~.96 is about as high as they go.


step 5: the final step is to format the data for me

all the formant frequency values (in hertz) need to be converted to synthedit pitch values. i have an app that does this one at a time, i could probably figure out how to automate it as it would take about an hour of tapping.

then the se pitch values need to be dropped a decimal place (eg. 5.5 becomes 0.55) and the data needs to be formatted:

(freq values for f1)
.55f, .59f, .37f, .37f....

(freq values for f2)
.55f, .59f, .37f, .37f....

et c.

then i'll work out the coefficients for the plosives and fricatives since that's proprietary.

so ultimately, a day or more of work.

[xoxos]

i've been sitting here for over five hours, my second long session today, making several unplanned improvements over the demo mp3 on timing, blending and consonants.

time for me to get up and stretch. i'm happy about what you'll be hearing soon!

[xoxos]

the joys of spending many hours alone with a computer...

i amended long a, i, o and u sounds to my diphthong list as the list i used to use only included 'ow' and 'oi' as diphthongs in received pronounciation (a fancy term for english english english).

originally, when i made syng1, i had amended long "i" to the list of diphthongs. my diphthong list ran "i" "ow" "oi."

when i added a, o and u i squeezed them in the list before i. in use i found this confusing, so i'm editing all my material to change the phoneme order.

one thing i have to do is change all the preset words in the patches, which are spelled out of phonemes..

..i had an "emotion" word set (used in the piece in the music cafe)

so i'm looking through the words for any diphthongs...

first word "dream." no diphthongs there.

second word "emotion." this is pronounced "e m o- sh u n."

except i had moved a o u i to the order of a i o u.

get it?

so the long o had been changed to a long i.

so it read "e m i- sh u n"

which isn't that funny on it's own, except it was situated below the word "dreams."

what a horrible story eh. sigh, the funny things i think up on my own.