From hearing a word to trying it
Google announced experimental pronunciation practice in Search on November 14, 2019. It used quokka as an example: instead of only playing a recording, a person could speak the word into a phone microphone and receive feedback.
The company described speech recognition separating the spoken input into sounds, with machine learning comparing the result against an expected pronunciation. Its example concerned identifying a particular sound in asterisk, rather than supplying only a single overall score.
The rollout had a language boundary
Google said the feature was rolling out for American English, with Spanish planned next. That was the initial announcement's language scope, not proof of immediate support for every language, region or accent.
The same post described adding images to dictionary and translation experiences, starting with nouns. These were separate additions intended to help people understand words. The announcement did not independently demonstrate that either feature produced a particular improvement in language learning.
Feedback is useful when its target is clear
Analysis: a practice tool should make it clear which pronunciation model it expects. A difference from that target is not, by itself, a judgment about whether a person's regional or cultural way of speaking is valid.
An evaluation could examine whether feedback identifies a specific, repeatable distinction and whether the learner understands what to try next. It should also include cases where the system mishears the input, rather than assume that every correction it suggests is warranted.
The 2019 feature proposal made Search more interactive by inviting a spoken attempt. Assessing its educational value would require evidence about that feedback and the learner's progress, not merely the presence of machine learning in the description.