One of the most ambitious Artificial Intelligence project is certainly the development of computers able to discuss with humans.
In the 70’s and 80’s, thanks to movies like "2001 a Space Odyssey" or "Star Wars", pictures about how we could be using talking machines started to spread through people’s imagination. Back to reality now, recently we have been the witnesses of several important technological progresses, including the historical victory of a computer over the best Go player, but also the first self-driving car, and the always increasing accuracy of automatic speech and face recognition.
To let machines talk with us in a natural manner, we understand that a fundamental step consists in converting speech in a set of symbols or text the machine can then process. This is exactly the research problematic of the field of Automatic Speech Recognition.
Authot is a French company offering an online speech-to-text service based-on the most modern Automatic Speech Recognition technologies.
An automatic speech-to-text system is basically a collection of computer codes designed to convert a speech recording into its word-for-word text version.
The text automatically generated may then be reviewed by annotators in order to remove repetitions, false starts, hesitations or syntax mistakes occurring frequently in spontaneous conversational speech. Transcripts may also be enriched with information such as music or noise.
That is the question we propose to develop throughout this summer in articles you will find on Authot’s website.
We will try to avoid too technical details and obscure mathematical equations, to share with you some insights regarding how can a machine recognize phonemes, the basic acoustic speech units composing words, and then propose the most likely text hypotheses given a speech utterance. Speech that the machine has of course never observed before.
Thank you to stay tuned and pending our next article, if you are interested in more details, we warmly invite you to have a look at this paper written by two pioneers of the field, tracing the history of several technological advances in "Automatic Speech Recognition – A Brief History of the Technology Development" published in 2006 in Encyclopedia of Language & Linguistics. This article can be easily found via a Google search.
Authôt: You speak. We write.