Problem: Speech to text apps are too good and correct lots of mistakes that learners would do well to notice

 

Solution: Students should listen to themselves while following the transcript to notice the corrections

Most speech to text apps or webpages correct many of the errors students make. So, I suggest students should listen to themselves while following the transcript to notice the corrections, before copying and pasting it into their favourite GenAI and asking for corrections. At the same time, they should correct any mistranscriptions like writing 'hungry' when the student meant to say 'angry'.

I like ChatGPT or Copilot as with either of them students can ask for them to display the original and the corrected versions in two columns of a table marking the errors and the corrections in bold. If students want to listen to it, they need to get it to display just the corrected version. If they do this using the microphone/headset they can hear it as it appears, or they can wait and touch the speaker icon.

If learners use their voice to ask for a more coherent version suitable for a learner at level B1 on the CEFR, for example, they can read and listen to it at the same time All this will help to make your students' repetition of the speaking task even better than it would be after simply doing the same thing again.

Here are three demonstrations of this technique using Soundtype.ai notta.ai & turboscribe.ai