Clone your own voice and have it read your lessons!

Text-to-speech is not a new service. Shortly after our debut at Thot in 1997, we offered this service on our pages, but for reasons of cost and number of uses, we abandoned it around the year 2000. Progress since then has made this technology much more accessible and useful for a variety of purposes.

Progress

The intelligibility of texts read by synthetic voices systematically came up against several obstacles, both technical and semantic, such as foreign terms, company names and abbreviations that were mispronounced, often rendering them incomprehensible.

Secondly, the quality of the texts, with their punctuation, grammar and spelling errors, often made them difficult to read.

Languages other than English were often poorly served, with voices that were robotic, atonal and lacking in diversity. Regional accents were never taken into account.

Finally, users had virtually no control over reading speed, intonation or style.

Thanks to artificial intelligence, all these difficulties are now overcome by the use of elaborate language models, multilingual dictionaries of previously unimaginable richness, and banks of contexts, accents and intonations that make text interpretation lively and adaptable to the content and the intended audience.

The Eleven Labs offer

We tested the Eleven Labs system, which is available in 29 languages, including all international languages and less common languages such as Arabic, Hindi, Malay and Filipino.

• Text reading (Text-to-speech)
  
  The same text can be read with different accents and styles. A vast library of voices and interpretation styles is available. A boxing match report, for example, can be read differently from a children's book or a report on a political debate.
  
  
• Voice cloning (Voice Lab)
  
  We can clone our own voice, and then use it as a reader. We can also create an artificial voice by modifying the parameters of the basic model. This technological breakthrough raises the question of the right to "image" - in this case, one's own voice.
  
  
• Translation and Dubbing
  
  Automatic translation of audio files is now relatively commonplace, but reinterpreting the audio tape in another language with the same voice style is a new service that makes it accessible to broadcast audio or video works in several languages, at low cost, such as training videos or educational podcasts.
  
  
  

We submitted this text and had it read by different voices. The interpretation is astonishing and the result impressive right from the start.
The generation may take a few minutes, but the result lives up to expectations.

A history of all your interpretations is kept and can be reused at will.
All files can be exported in mp3 format.
APIs are also available for systematic and intensive use.

Some players are only available in version v1. In this version, if numbers are written in numerals, they will be recited in English. To have them in your language, they must be written in text: for example, "twenty-nine" instead of "29". If you take a voice from version 1 with version 2, you will observe a significant variation in style, accent or volume which is not representative of the level of quality possible with version 2.

Don't forget to select your starting language and choose a reader that matches the style of your text.

Eleven Labs speech synthesis

Illustration: Nataly-Nete - DepositPhotos