Hidden code inside the Android version of Google Translate points to a pronunciation-training feature in active development: a “Practice” mode that lets users listen to a translated phrase, record their own attempt, receive a score, and get corrective feedback.
The feature is not visible to users yet, and there is no official confirmation or launch date.
The gap this would close is a familiar one. Google Translate has long been able to tell users what to say, but not whether they said it clearly enough to be understood. A pronunciation-focused loop would push the app further from one-off translation and closer to guided speaking practice.
What it does
Users can listen to a voice sample of their translated phrase and review several alternative phrasings before attempting anything. Tapping a “Pronounce” button opens recording, and Translate analyzes the attempt, returning an overall score with suggestions for improvement, Android Authority reported. The loop is deliberate: listen, attempt, evaluate, retry.
For learners who find audio samples alone insufficient, the feature also appears to offer phonetic guidance in plain written text. It skips IPA, or the International Phonetic Alphabet, in favor of something a typical learner can read more easily. Android Authority also said the app asks users to try again if the recording is too unclear to evaluate, rather than forcing a score from bad audio.
The “Practice” mode sits alongside two other AI features in development inside Translate, “Understand” and “Ask,” which Android Authority has tracked across multiple builds. That points to a broader interactive layer inside Translate rather than a single isolated feature.
Why it could ship
The technical ground here is not entirely new. In August 2025, Google announced AI-powered language-learning tools in Translateincluding conversation practice that generates speaking and listening exercises around a user’s goals. Google’s support pages now show that “Practice” is already in limited beta for some language pairs and markets, with AI-generated speaking and listening activities in the app.
The feature shown looks narrower than a full language-learning mode. The evidence so far suggests a narrower tool aimed at helping users produce more recognizable speech from translated phrases, not a full accent-coaching system or a replacement for broader language learning.
The practical value will depend heavily on language coverage, scoring quality, and rollout scope. Android Authority said Spanish and English were the only language pairs confirmed to work in the teardown, and the full list of supported languages is still unknown. Google has followed a similar incremental pattern elsewhere, including its recent overhaul of Android sideloading security, and the company is moving AI deeper into core products, visible in how Google’s AI is reshaping search results and publisher content. If this feature ships, it would fit that broader push to make AI guidance a more visible part of everyday tools.
Translate helping users understand a phrase has never been the hard part. Helping them say it clearly enough to be understood is harder and more valuable. The teardown shows Google is working on that problem. Whether the final feature is broadly useful or narrowly limited is still an open question.
Also read: Nvidia’s new Blackwell Ultra and Rubin roadmap show how fast AI infrastructure demands are escalating.