Google has launched a beta app called Project Relate, inviting people with speech impairments to use a voice assistant while simultaneously contributing to research that could make speech recognition more inclusive. The program addresses a long-standing gap: AI-powered voice tools have historically performed poorly for users whose speech patterns differ from the training data used to build these systems.
The short version
- Project Relate beta launched; sign-ups at g.co/ProjectRelate
- Requires 18+, Google account, Android phone (OS 8+)
- Currently available to English speakers in the US, Canada, Australia, and New Zealand
- Volunteers record 500 sentences, roughly 30 to 90 minutes of work
- Three app features unlock after contributing voice samples
What Relate Actually Does
Participants who complete the recording process get access to three core capabilities. First, real-time transcription displays what the user is saying as they speak. Second, a "Repeat" feature plays back the user's words in a synthesized voice optimized for clarity, useful during conversations or when speaking to smart devices that otherwise struggle to understand. Third, the app connects directly to Google Assistant, allowing users to control smart home hardware through their own natural speech rather than adapting to what the system expects.
The practical impact is straightforward: for someone whose speech patterns fall outside typical parameters, these features turn voice control from a frustrating failure loop into something that actually works.
Why This Gap Existed
Without sufficient training data representing diverse speech patterns, Google Translate and Google Assistant remained largely inaccessible to users with conditions like ALS, traumatic brain injury, or Parkinson's disease. The underlying issue is a classic data problem: AI systems reflect the data they learn from. If that data skews heavily toward certain dialects, cadences, and articulation patterns, everything else gets filtered out.
Google launched Project Euphonia in 2019 specifically to begin addressing this. That initiative collects voice samples from patients with speech-affecting conditions under controlled conditions. Project Relate represents a different approach: rather than Google sourcing participants through clinical partnerships, it opens the process to anyone who meets basic criteria and wants to participate directly. The trade-off is less controlled recording conditions but broader reach.
Who Should Consider Signing Up
This is most relevant for adults experiencing speech changes due to neurological conditions, progressive diseases, or injury who find existing voice tools unreliable. You need an Android device and the ability to complete a recording session. The 500-sentence requirement is substantial, but the access granted afterward provides immediate utility rather than asking participants to contribute for abstract future benefit.
If you already use Google Assistant successfully and are understood clearly by it, Project Relate offers less immediate value. The research angle benefits the broader community, but the practical tools are most meaningful for those currently excluded from voice AI.
Our Analysis
This initiative highlights a fundamental tension in AI development: systems improve fastest when trained on consistent, abundant data, which tends to reward the majority use case. Users whose needs are most acute—those who would benefit enormously from accessible voice control—often have the least representation in training sets.
Project Relate attempts to flip that equation by creating genuine value for participants rather than just asking for altruistic data donation. That model is more sustainable than outreach that asks marginalized users to do unpaid work. The limitation to English and Android is understandable for a beta but will need addressing before claims about meaningful accessibility can be fully credited.
The bigger question is whether Amazon and Apple follow. Voice assistants from those companies face the same training data gaps. A rising tide approach—industry-wide initiatives sharing anonymized datasets—could accelerate progress faster than parallel single-company efforts.
What to Watch Next
Monitor whether Google expands Relate to iOS and additional languages. Watch for similar announcements from competing platforms. Track whether the voice data collected translates to measurable improvements in existing products like Google Translate, which currently struggles with impaired speech input. Real-world accessibility gains matter more than beta program size.
Key takeaways
- Voice AI has historically underperformed for users with non-standard speech patterns due to training data gaps
- Project Relate gives immediate practical tools to participants while building a more representative dataset
- The initiative models a participant-first approach to inclusive AI development
- Expansion beyond English and Android will determine whether this achieves broad accessibility impact
- Competing platforms likely face similar gaps and may announce comparable programs
Frequently asked questions
What devices and operating systems support Project Relate?
Currently only Android phones running OS 8 or later. IOS is not supported at launch.
What speech-affecting conditions does Project Relate aim to help?
The program focuses on users with neurological conditions that impact speech clarity, including ALS, traumatic brain injury, Parkinson's disease, and progressive conditions like muscular dystrophy.
How long does the initial data contribution take?
Volunteers record 500 sentences, which Google estimates takes between 30 and 90 minutes depending on speech pace and condition.
Is this available outside English-speaking countries?
Not yet. The beta is limited to users in the United States, Canada, Australia, and New Zealand who speak English.
Does participating guarantee improvements to how Google products work for my specific speech patterns?
The app provides personalized features that adapt to your voice immediately. Broader improvements to Google Assistant and other products depend on aggregate data contributing to model updates over time.