3

There are many text to speech software, but the one I'm looking for would allow to synthesise the user's own voice. And possibly imitate the voice of someone else without consent (just for joking, kind of what is done with photoshop today on images).

| improve this question | | | | |
  • 1
    What is your budget? – Nicolas Raoul Jun 9 '14 at 3:26
  • @NicolasRaoul : Don't care... I want to see first if it exist. – user2284570 Oct 23 '14 at 3:39
  • There are now several "voice cloning" apps that can mimic voices from recordings. – Anderson Green Nov 27 '19 at 16:16
3

As far as I know text-to-speech based on the user's own voice is still an open research question in speech synthesis. VocaliD was recently presented at a TEDWomen Talk by Rupal Patel:

VocaliD creates custom crafted synthetic voices by combining the recipient’s residual vocal abilities with an anatomically similar voice donor’s speech database. The result is a voice that sounds like the recipient in age, personality and vocal identity but is as clear and understandable as the donor’s speech.

TEDWomen Talk: "Synthetic voices, as unique as fingerprints" (Dec 2013). Abstract:

Many of those with severe speech disorders use a computerized device to communicate. Yet they choose between only a few voice options. That's why Stephen Hawking has an American accent, and why many people end up with the same voice, often to incongruous effect. Speech scientist Rupal Patel wanted to do something about this, and in this wonderful talk she shares her work to engineer unique voices for the voiceless.

| improve this answer | | | | |
  • In 2000, I saw a linux software for this : it was English only with many many cursors on the screen, and I didn't read his name (I was still learning how to read at that time)... It allowed to synthesise many voices : but they were no mechanism for own voice's text to speech.. – user2284570 Jun 8 '14 at 1:40
  • 1
    lyrebird.ai – user2284570 Jun 4 '18 at 18:20
2

General information

Although recording own voices is done since ~2004, it is still a hard task. The Nemours Speech Research Laboratory is looking for volunteers to record voices for voiceless children. They record 1600 sentences to build a voice. This is needs a significant amount of time, some practice and good equipment, otherwise your breath and background noise will be recorded as well.

The Edinburgh University is using HTS (BSD like license) which is based on HTK (proprietary license) and Festival. The Festvox project of Carnegie Mellon University is closely related. They have built voices for that system.

In the following you'll find a few voice recording services. I never tried them myself, because I considered them as too expensive for the use I had in mind.

Acapella

Acapella offers custom voice creation. It will be very expensive though and you'll need about 3 months to get take all the recordings (for high quality). Still, the custom voices are not as good as their regular ones.

See also the demo site, the FAQ and the Eric and Garmt recordings vs. synthesis.

CereVoiceMe

CereVoiceMe is the only service I know of that gives a price in advance: For about 775 USD you can record "a couple of hours" of your voice and it'll be converted to a TTS voice. I'm not sure if they offer French voice recording.

| improve this answer | | | | |
  • I used to remember a software with lot of cursors for switching parameters when I was a kid. Looks my mind was wrong. it‘s far from be able to imitate the voice of someone else without consent. It‘s also far from being short and “Do It Yourself”. I now figure this can require true artificial intelligence (but not at the human brain level since parrots can do it (pronounce words they never heard with a particular voice) ) – user2284570 Nov 17 '15 at 21:29
  • @user2284570: brains are awesome :-) Maybe it'll take 10 more years. The progress from 2004 to 2015 is promising. There are really good TTS synthetical voices available. Much better than Microsoft Sam... – Thomas Weller Nov 17 '15 at 21:36
  • I also realize how it can be harmfull as you would be able create fake proofs. Or at the reverse (it‘s difficult to get proof that some persons wanted to create a terrorist attack before it happened) make a fake call with the voice of a terrorist in team to direct an attack to a safe place were police wait them. – user2284570 Nov 17 '15 at 21:42
  • @user2284570: As you say you recall a software with lots of sliders etc, that would better match this question – Thomas Weller Nov 17 '15 at 21:43
  • Yes, finally, recognizing and proofing the identity of a voice will be as hard as identifying a photoshopped picture today. – Thomas Weller Nov 17 '15 at 21:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.