~250 million people worldwide can’t rely on speech technology today due to difficulty speaking.

In 2018, we realized that our speech recognition technology could be improved to help people with speech impairments. Google’s Speech and Research teams worked together for several years to analyze more than 1 million speech samples recorded by over 1,000 people, before developing Project Relate, a beta Android app. As an early stage technology, we are seeking trusted testers to try out and provide feedback on the app at g.co/ProjectRelate so that everyone can successfully use speech technology.

In order for Project Relate to become a reality, we first needed to build a model using lots of speakers, different types and severities of speech impairments, and acoustic conditions. Doing so then allowed us to build customized models for individual users, once provided with several hundred phrases of that person’s speech. User feedback enables us to make continuous improvements while we are still in beta, and working only with English speakers aged 18+ in specific countries. Project Relate is a free application and available via Android. In the research that preceded its development, we offered gift cards to adult participants interested in contributing speech samples to improve Automated Speech Recognition (ASR).

For millions of people, being able to speak and be understood can be challenging as a result of speech impairments. Voice-enabled tech is becoming increasingly more ubiquitous, but speech recognition algorithms don’t always work as well for people with atypical speech. Project Relate is an Android beta app that uses machine learning to help individuals with speech impairments communicate more easily with others and interact with the Google Assistant. By personalizing its speech recognition to the individual, Project Relate is a new technology that seeks to improve everyday life for people with speech impairments.

*2018 - 2021: Google Speech and Research teams collected more than 1 million utterances from more than 1000 people.

*2019: Using the research above, our first prototype was created.

*2020: 2 Google employees with speech impairments tested the app and provided feedback to product and engineering teams.

*Nov 2021: We invited our first users.

*2022: We scaled to 5 countries.

By providing the app with examples of you talking, you can train Project Relate to understand your individual speech patterns. Once you do so, you receive a custom speech model, and unlock access to 3 features.

With the Listen feature, you can transcribe speech to text in the Relate app so you can use your voice to write docs or text messages. With the Repeat feature, you can repeat what you’ve said in a clear, computerized voice, for greater ease in face to face communication. With the Assistant feature, you can speak directly to your Google Assistant so you can take care of tasks such as setting a timer, playing a song, or asking for directions.

Project Relate runs on Android, and we are currently in beta. We’re seeking English speakers 18+ in the United States, Canada, India, Australia, and New Zealand interested in trying the app at g.co/ProjectRelate.

Every Project Relate user receives a personalized model. While our app is currently by invitation only, we continue to scale the training and delivery of personalized models for users at a measured pace; our goal is to offer Project Relate as a free communication tool for people with speech impairments at scale.

Project Relate is currently available as a beta in 5 countries,, and significant improvements to word-error-rate (WER). We plan to scale to more countries and more languages in the future, in order to help more people benefit from speech technology.