Who hasn’t enjoyed having Google Translate’s robotic voice read sentences to mock his awkward accent? Over the years, this peculiar sound has become a legend. Some have even managed to divert artistically, like rapper Orelsan, with “Computer Assisted Freestyle” (available on YouTube). However, with the disappearance of this symbolic sound, a new page is about to open. The days of Google Translate and Google Assistant with funny voices will soon be over.
Google Assistant: The new Tacotron 2 expresses voice better than a human
Google has announced a new voice produced by artificial intelligence. This voice, called Tacotron 2, is almost indistinguishable from the human voice. The system is based on two artificial neural networks. The first network converts a text into a spectrogram, that is, a visual representation of sound frequencies. This spectrogram is then transmitted to WaveNet, a system developed by Google DeepMind that can read this representation to create an audio file.
Here are the different excerpts published by Google: just stunning. Each sentence is read by an employee of the company and Tacotron 2. It is very difficult, if not impossible, to know whether the sentence was read by a robot or a human. Tacotron 2 can change its pronunciation according to punctuation and even pronounce the most complex names.
This technology is not meant to stay in the lab: it will soon replace the existing voice of Google Assistant and Google Translate. But for now, the AI has been trained to mimic only a single female English-speaking voice. It will take longer training to master several languages and speak like a man. Meanwhile, discover this new sound for yourself by reading complex sentences and comparing it to a human voice.
Support our work ❤️
If you enjoyed this article, consider leaving a tip to help us keep publishing great content.
