French voice artificial intelligence is looking for voices

In lack of sounds to train their artificial intelligence (AI) models, French companies and labs of vocal techniques ("voicetech") will launch a campaign to ask French speakers to give a little of their voice, explained Karel Bourgois , president of the Voice Lab.

About thirty players in the sector have pooled their strength in this association to bring together thousands of hours of recorded voices, essential to feed and improve models of vocal artificial intelligence.

To discover

• Crosswords, arrow words, 7 Letters... Free to play anywhere, anytime with the Le Figaro Games app

“Together we put together 9000 hours.

But we are start-ups and SMEs facing giants like Microsoft or Google, which with YouTube has millions of hours.

In France, +datasets+ are few in number and often without a license for commercial use, hence the difficulty in training artificial intelligence.

Recently, a young researcher spent two years just compiling her data

,” lamented the entrepreneur.

To go further, the Voice Lab, in partnership with the Mozilla Foundation, will help relaunch the collection of voices in French on the

site , where everyone can register by reading a text.

And, in September, he will launch a campaign for a new version of this tool,

.

"Automate call centers"

Another track, with the Human-Num laboratory, the “listening to speak” project: a truck that travels through France to record voices, more diverse than the voices of radios or TVs.

The Voice Lab is also discussing with Radio France, France Télévisions and the INA, but comes up against the legal vagueness concerning the notion of use for training purposes of artificial intelligence.

In 2021, the Voice Lab won a public call for projects and obtained 4.7 million euros over 5 years to pool voice data, create common models, expose the services of its members, for research purposes or commercial.

Read alsoDo you speak the language of the metaverse?

A booming sector, “voicetech” includes voice recognition and synthesis, emotion analysis, speaker identification, oral transcription of texts, elimination of accents or even voice imitation and transformation, including in real time.

These techniques are of interest to the general public as well as large groups who want to use voice as an identifier or to automate call centers.

In January, Microsoft presented VALL-E, an AI model that can imitate a voice from 3 seconds of recording.