Date   
[espeak-ng:master] reported: Max WPM on android app too low #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #650 Max WPM on android app too low
By thgcode:

You can change the WPM even further by using the standard Android TTS settings. The speed set in eSpeak is multiplied by the speed set in the Android text to speech settings. Sonic is included in the android build.

[espeak-ng:master] reported: Reverse-engineering espeak-ng for STT #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #669 Reverse-engineering espeak-ng for STT
By guest271314:

@materia-cccp The concept is simple. Since espeak-ng is currently capable if mapping text to audio and given the audio output is the same or similar when the same input text and voice is used it should be possible to adjust the existing code to map the audio that will be produced by input text to output text.

[espeak-ng:master] reported: problem with the testing languages #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #662 problem with the testing languages
By tc-dl:

I switched to ubuntu 16.4 there it worked. but there is no PCAudioLib it what is needed? I hope after you edit the dictionary files will work.

[espeak-ng:master] reported: Reverse-engineering espeak-ng for STT #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #669 Reverse-engineering espeak-ng for STT
By materia-cccp:

reversing function is not reverse engineering its a whole nother task

[espeak-ng:master] new issue: Reverse-engineering espeak-ng for STT #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Issue Created by guest271314:
#669 Reverse-engineering espeak-ng for STT

espeak-ng can produce .wav files using --stdout.

What is the relative difficulty of reverse-engineering espeak-ng to take input as .wav file produced by espeak-ng to analyze the audio and output plain text?

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

Sure. You could also clone the repository and make the changes in your cloned version of espeak.

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By destroyersboy:

Hey, I need some scrutiny.

I'm gonna talk to people with more Turkish grammar and emphasis. We don't want to do anything wrong, do we?

We can add changes to a new branch. When the tests are over, we get them to the master.

[espeak-ng:master] reported: espeak-ng sapi5 #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #633 espeak-ng sapi5
By burakyuksek:

I agree.

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

So this is fixed now?

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By destroyersboy:

At the moment there seems to be no problem with Turkish synthesis. I compiled immediately after the change.

Moreover, in the pronunciation of double letters, there was no contradiction to the Turkish language rules and emphasis.

Updates to Github #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

1 New Commit:

[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
07012f607360: tr: use long consonants for gemminated plosives, like in Italian.

Modified: dictsource/tr_rules


[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

I've made the change to use the long consonants, like in the Italian gemminated consonants. For 'etti', this sounds very close to the sound that Yelda makes in the Nuance voice, and there is a difference between 'eti' and 'etti'.

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

If the sound go back to 1.48.03, then the issue reported in #152 would be present. I would like to modify it so that both this and #152 are fixed.

Notice that eSpeak sounds different to Yelda in with eSpeak you can clearly identify the consonant in both cases, but with Yelda the sound is much more subtle. That is what #152 was about, hence my reluctance to revert to the 1.48.03 behaviour without also fixing #152. As such, I would want to make both changes together.

I'm curious why you don't like the first version I listed, as using that in 'etti' sounds similar to the way Yelda is pronouncing it. That is also what the Italian eSpeak voice is doing for the geminated consonants.

Details

It sounds like there is a glottal stop (or similar) for the first t or other consonant sound. It is like how in English you would pronounce something like 'set to', 'lamp post', or 'guard dog' in normal speech.

Experimenting with trying to produce those sounds compared to a glottal stop, it looks as if the following are happening with the first geminated consonant (and other unreleased stops): 1. the tongue is in the position of the stop being produced, not the position of a glottal stop; 2. the glottis is closed to prevent air escaping; 3. the mouth is closed in the p sound (like it is in the regular version before the release part of that sound); 4. the initial closure phase is held a fraction longer in the geminated version of a consonant than the non-geminated version. NOTE: This also applies for the nasal consonants like n (e.g. in the Italian 'anno') where the tongue is held at the roof of the mouth for a fraction longer than the non-geminated version.

I'm not sure how to achieve this in eSpeak, so some experimentation would be needed. I'm also not sure if a regular glottal stop would be sufficient in this instant. The eSpeak glottal stop is not great, so would need improving.

I'm also not sure if code changes (in addition to pronunciation rule changes) would be needed.

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By destroyersboy:

Hi @rhdunn again,

Frankly, I don't know exactly what you want to do. If you want to say that the pronunciation I'm talking about is incomplete and wrong, it is necessary to listen to Yelda, Nuance's Turkish voice. https://www.nuance.com/omni-channel-customer-engagement/voice-and-ivr/text-to-speech.html#! please write: "etti, appe, eddard"

Will you go back to 1.48.03?

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

The ones you selected were the ones that @burakyuksek had an issue with in #152.

It would be better then to look at implementing the unreleased stop behaviour I mentioned above (like the English voice is doing for different consonant pairs in things like doctor) for the first of a geminated consonant pair and revert the doubled consonant change from #229 and #152

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By destroyersboy:

Hello @rhdunn , I think we have found the solution. espeak-ng -v tr "[[s'ete]], [[s'ette]]" and espeak-ng -xv tr "[[Ed'aRd]], [[Edd'aRd]]"

As @burakyuksek said, words like 'eddard' are broken. More settings can be made for Turkish. But as it stands, it looks better.

I have friends who listen to books and write long articles with Espeak (15-20), I'm not happy with Espeak reading like a one letter.
The thinks of my friends support my opinion.

I will work on Espeak's syntax. In the future, I'il help you with more.

I think that's the solution, we have to implement it.

Updates to Github #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

For 'd' and 'dd', does

espeak-ng -xv tr "[[Ed'aRd]], [[Ed:'aRd]]"

or:

espeak-ng -xv tr "[[Ed'aRd]], [[Edd'aRd]]"

sound better?

For 'sete sette' (made up), does:

espeak-ng -v tr "[[s'ete]], [[s'et:e]]"

or:

espeak-ng -v tr "[[s'ete]], [[s'ette]]"

sound better?


[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

For 'd' and 'dd', does

espeak-ng -xv tr "[[Ed'aRd]], [[Ed:'aRd]]"

or:

espeak-ng -xv tr "[[Ed'aRd]], [[Edd'aRd]]"

sound better?

For 't' and 'tt', does:

espeak-ng -v tr "[[s'ete]], [[s'et:e]]"

or:

espeak-ng -v tr "[[s'ete]], [[s'ette]]"

sound better?


[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By burakyuksek:

Hello, I think espeak sounds fine the way it is, because for example when saying "ettim", we don't say 2 t's, we make the first t silent and say the second one. Of course when this first t is silenced, there is a slight pause. That's why I suggested that change.

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By destroyersboy:

In #152 (released in 1.49.1 -- 2017-01-21), the double letters tt, bb, cc, çç, dd, kk, and pp were modified to use _::p, etc.

I don't fully understand the purpose of the change here. Q: kyu, w: dabılyu and x: iks. That feedback is true.

I do not have the technical knowledge to interpret other technical parts and understand Espeak's phonetic syntax.

On the other hand, during fast reading and slow reading, double letters is emphasis and double, not read as a single letter. In the present Espeak, the double letters found above, are passed as singly. It should be read and emphased in pairs.

In #229 (released in 1.49.2 -- 2017-09-24), the long pause was modified to work better at higher speech rates. This (da3d534) should really be reverted and the code should be modified to better support higher speech rates, as there are other artefacts in all languages in this case.

@Burakyuksek said the error occurred as a result of the change. The revision is reverted to version 1.48.03. Because there's no problem here.

Not only at high speeds, but at slow speeds, exactly what the word is.

We can better understand this phonetic subject with the word 'eddard'. In words where the letter D is double, the tension of the letter should not be lost. This word is now read as 'edard'. Letters like D and T should be emphasized more than this level.

  1. Does 1.49.1 sound wrong as well? Yes, Wrong:
  2. Words that repeat a silent letter twice has the first instance of the letter muted. For example, in the word gitti, the first t needs to be muted.

The right lines: > 2. The letters x, q and w are not pronounced correctly. X needs to be pronounced as iks rather than eks, w should be pronounced as çift v rather than duble v because duble is not a Turkish word, and the letter q needs to be pronounced as kyu rather than kve.

  1. Is this a dialect difference between you and @burakyuksek (the person who raised #152)?
    No, no dialect difference. All Turkey, Istanbul Turkish speaks. This is the official.

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By rhdunn:

In #152 (released in 1.49.1 -- 2017-01-21), the double letters tt, bb, cc, çç, dd, kk, and pp were modified to use _::p, etc.

In #229 (released in 1.49.2 -- 2017-09-24), the long pause was modified to work better at higher speech rates. This (da3d534438398840af0f155fe3b01c630f9d34c5) should really be reverted and the code should be modified to better support higher speech rates, as there are other artefacts in all languages in this case.

I have some questions: 1. Does 1.49.1 sound wrong as well? 2. Is this a dialect difference between you and @burakyuksek (the person who raised #152)?

This is a process called Gemination. In the Turkish section, wikipedia transcribes the geminated consonants with a long consonant. That is:

espeak-ng -xv tr "[[et:im]]"

This is how the Italian (it) voice handles gemination for the plosives. In that example, eSpeak still has a pause in the above before the t, but the pause is a lot shorter than the #152 version:

espeak-ng -xv tr "[[e_::tim]]"

The original version:

espeak-ng -xv tr "[[ettim]]"

sounds as if it has too much release on the first t (the exhilation of air after the initial closed part of the sound).

It might make sense to use the original tt variant, but modify the phoneme rules to make the first t an unreleased stop making the geminated consonants [t̚t]. This is audible in the en voice for distinct consonants, like in the words 'apt' and 'doctor'.

@destroyersboy and @burakyuksek which of these is best for the both of you? Is using [[t:]], etc. acceptable?

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By destroyersboy:

Yeah, there's nothing wrong with the word 'elli'. I must have accidentally added him as an example.

According to my tests: tt, dd, cc, bb, çç, dd, kk, pp...

[espeak-ng:master] reported: (Turkish) vocalization of words with double letters saying of one letters #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #668 (Turkish) vocalization of words with double letters saying of one letters
By hozosch:

When I test "elli", I can't tell a difference in the both version. I get though what you mean now. Double t and stuff should be pronounced as two t consonants, instead of making a pause between them.