Google develops an AI tool that helps people with speech deficiency

Google develops an AI tool that helps people with speech deficiency

credit: third party image reference

The Google team is working on a new project that aims to help users with speech problems. 

Taking into account the role played by virtual assistants and voice recognition services, they are analyzing how to use this dynamic to improve the communication and interaction of these users. 

So far, people with speech problems are excluded from these technologies since the AI has not been trained to contemplate different scenarios. 

credit: third party image reference

To solve this problem, Google developed Parrotron: 

Parrotron, an end-to-end voice-to-speech model that maps an input spectrogram directly to another spectrogram, without using any intermediate discrete representation [...] We demonstrate that this model can be trained to normalize the speech of any speaker, regardless of the accent, prosody and background noise.

Following this dynamic, this neural network can help people with speech disorders (as a consequence of different diseases) to communicate and be understood by other people as well as by assistants or voice recognition using the devices. 

credit: third party image reference

The Google team tested this technology with different users, and the Parrotron system succeeded in reducing the word error rate from 89 to 32%. A more than promising result that shows the potential of this tool. 

We can read all the steps of these tests and the technical details of the training in the report shared by Google.