Asking permission for running TTS experiments with LV data

[giacomos]

Dear Librivox users,

I am Giacomo Sommavilla, I work on Speech Synthesis for an Italian
Public Research Institution (ISTC-CNR, http://www.pd.istc.cnr.it) and also
for a small new company.

Speech synthesis relies heavily on data (i.e. recordings of people
reading books), and this is how we came across Librivox. It is a
really wonderful resource!

We would like to get in touch with Librivox readers and to ask them
permission to use their voice recordings to run speech synthesis
experiments and create digital voices.

A student and I are experimenting with some spanish recordings to
create synthetic voices for the Spanish Language.

The readers that we would like to contact are

| forum username | librivox username |
|----------------+-------------------|
| tux753 | tux |
| mongope | Mongope |

I tried to contact the readers through personal messages of the forum
but I didn't find a way to do it.

Can you tell me how to get in touch with them?

Thank you very much and best regards,

Giacomo Sommavilla

[TriciaG]

Our recordings are all put into the public domain, which means that anyone can do anything that they like with them. However, it's considerate to ask for permission as well.

Both these readers are active. You, as a brand new member, cannot use the private messaging system. I will send them a PM and point them to this post, so they can respond to you.

[giacomos]

Thank you very much Tricia, I have just received a positive Private Message from a reader, but unfortunately I cannot reply to him, since I still don't have permission to write PMs.

I thank him as well in this thread for his kindness!

[Availle]

I think you will need to have one more post in the forums to be able to write/reply to private messages. It's our anti spam measure

[TriciaG]

Via PM:

Ok, Tricia!! No soy más explícita, tengo problemas con el PC.
Me parece bien, estupendo.
Saludos.
Mongope

[giacomos]

Thank you very much Tux and Mogope!

If someone of you is curious about TTS and our experiments, I can
suggest you to try out our demo available at this page

http://fic2fatts.tts.mivoq.it/

There are several voices that we have built with our software (which
is a modified version of Mary TTS).

One of the biggest problems we encounter with speech synthesis is the
fact that digital voices have a flat and unnatural prosody, especially
on long sentences. One of the reasons for this problem is that
normally, to create digital voices, relatively short sentences are
used, which are disconnected from each other, and are chosen in such a
way as to best cover the contexts of a language. For this reason,
these sentences may also be difficult to read.

On the other hand, one of the synthetic voices found on our demo page
above is "istc-speaker_internazionale", which was built with long,
articulate recordings (from the syntactic point of view and with
dependent clauses). Moreover, the sentences were drawn from texts made
up of lengthy meaningful paragraphs. The
"istc-speaker_internazionale" voice is very good in terms of
intonation.

Since librivox recordings is similar to this material, we would like
to use it to understand if we can generate digital voices that can
read long texts more appropriately and naturally.

I hope you find this topic interesting.

Thanks again and best regards,
Giacomo


edit: I forgot to say that "istc-speaker_internazionale" is an
Italian synthetic voice.

[VfkaBT]

Question for Tricia, not the original poster: what if our voices are used for illegal purposes, like phone scams and certain kinds of pornography? Is this a new frontier for the legal community?

[TriciaG]

You'd have to ask a lawyer or a legal researcher, but as far as I know, no one has been prosecuted for their voice being used for nefarious purposes without their knowledge.