Google’s WaveNet Mimics a Human’s Voice

Google DeepMind researchers have been busy creating a new artificial intelligence called WaveNet that has the ability to mimic human speech like no other machine out there.  The machine learns to mimic individual sound waves made by humans and is currently being tested in both US English and Mandarin Chinese, and so far, the results are pretty astonishing.  Although it hasn’t yet achieved 100 percent perfection, it is pretty close.

The way in which WaveNet works is through speech synthesis, more commonly known as text-to-speech (TTS).  Because WaveNet is an AI, it also has the ability to learn and teach itself and is classed as a convolutional neural network.  It can carry out deep learning techniques which mean after been trained on different types of data it can begin to make inferences about new data.

Google DeepMind’s team are so confident in their system that they have published samples that you can listen to online to decide for yourself how human you think they sound.  They have also applied WaveNet to speech recognition and the researchers commented, “We trained WaveNet with two loss teams, one to predict the next sample and one to classify the frame, the model generalized better than with a single loss and achieved 18.8 PER on the test set, which is to our knowledge the best score obtained from a model trained directly on raw audio on TIMIT.”  So, look out the world, there may be talking, human-sounding robots coming sooner than you think!

---

More News To Read 

• Apple Get One Up On Galaxy By Doing…. Nothing
• Renewables.ninja – the New Super Cool Tool That Calculates Renewable Energy Output
• Solar Panels Just Got Twice as Efficient
• The Truth About Baking Soda
• Scientists Get Up Close and Personal With Jupiter