The podcast about a 5-step prompt for students to use to get feedback on their speaking

I chose 10 short recordings made by my pre-intermediate and intermediate students when I was a teacher and got verbatim transcripts using Rev.com 

Perhaps as a result of selecting from only recordings of under one minute, there are more PINT (pre-intermediate) recordings that INT (intermediate) ones. 7 x PINT and 3 x INT

I then made screen recordings on my Android phone of me commenting on the recordings and playing them as I outlined the ideas behind the technique that I'll be talking about at various conferences this year.

You can watch them here They vary in length from under two minutes to just over seven minutes. Here they are in the order I recorded them:

Some were recorded at home (4), but more (6) were recorded in class with the noise of five or six other people speaking in the background. None the less, the transcriptions didn't contain too many mistranscriptions caused by faulty pronunciation or the use of unfamiliar proper nouns.

When looking at my whole collection of student recordings, there are many more recorded in class than recorded at home as everyone recorded themselves at least once in class every day and despite my efforts to persuade students to record themselves at home very few of them did so.

Almost inevitably nowadays, I thought it might work well to upload all10 videos to NotebookLM and I was pleasantly surprised that mp4 files were as acceptable as mp3s.

I used this customisation:

This is a series of short videos based on recordings by Pre-intermediate and Intermediate students made in class or at home. Each video elaborates on Chris Fry's ideas about how students can record themselves, get transcripts and then seek help from ChatGPT using the long 5-step prompt he is developing.

Please concentrate on the overall concept rather than the contents of the students' recordings although you can mention extracts that illustrate the main points of the process.

The resulting podcast was very flattering and I was amazed with how well NotebookLM collected the different comments I made about each of the videos and made a coherent account out of it.

This is a link to the NotebookLM audio overview but I really prefer playing it with a synchronised transcript using Rev.com , which you can do by clicking on the link below.

Transcription of podcast about recordings using Turboscribe ai and ChatGPT to help students boost their speaking

A Comparison of how ChatGPT, GSE Text Analyzer, Text Inspector and Write&Improve evaluated a series of texts

 

My ideas about how students can use transcripts produced by tools like Turboscribe.ai and Rev.com combined with ChatGPT to get suggestions for ways to upgrade their language depend on ChatGPT grading their original transcript correctly and then producing progressively more complex versions of it for them to read, listen to, study and take notes on. They would then repeat the speaking task with a different listener, hopefully benefitting from the exposure to the emergent language.

So, I have tried examining how ChatGPT's grading compares with other tools that claim to do the same. The chart above is based on only one B1 student's original transcript. There are vast differences between the four different tools.

To convert the CEFR scale to numbers I used this conversion chart:

It must be said that both Text Inspector and Write&Improve only claim to evaluate written English.

As there is a lot of data packed into the chart, I decided to ask ChatGPT to compare the accuracy of the four. Am I being cynical when I say, as expected, ChatGPT rated itself to be one of the most accurate?

The conversation with ChatGPT on the subject is far too long to include here , but if you are interested you can read a summary of it here.     

This is the conclusion, which means I am reassured that ChatGPT is doing a 'good enough' job of rating and producing progressively more difficult versions.