ChatGPT interviewing me about the Spanish Couple on the Titanic

Daniel Martin suggested on Facebook that with the right prompt ChatGPT could interveiw anyone using voice. I adapted his prompt to make it easier for students to use it with any text the wanted.

Here is the text that I used for this experiment:
https://www.bulldogz.org/post/titanic-tales-a-single-spanish-victim

Here is the ChatGPT conversation, which is only a text chat:
https://chatgpt.com/share/67b0f234-9a28-800b-9000-c3f06260a3ff

And here is the recording I made of the interview:
https://www.rev.com/app/transcript/NjdhZmEyZDg1YWFjZGMzOTA2YjQ1ZWY1N3dLMk9scE11b2NY/o/VEMwMTk3MzQ5MDA2

You can watch a video about it here:

I would love to see a similar video made by a student.

Turboscribe versus Rev - the two best and most generous transcription tools

I asked NotebookLM to do the heavy lifting of comparing three transcripts saved from Turboscribe (x1) and Rev (x2)

This is an edited version of how it replied to my question:

“I am interested in any differences between the three transcripts particularly regarding the use of interjections and the use of phrases like 'like', 'you know', 'um' etc”

Differences appear in the transcripts regarding interjections and phrases such as "like," "you know," and "um." The transcript generated by TurboScribe.ai, includes these kinds of interjections and phrases more frequently compared to the two transcripts generated by Rev.com

For example, the TurboScribe transcript includes the parts in bold that are missing from the Rev transcripts:

·         "[M] Right. It's like we're on the edge of this revolution, you know, and how we learn and practise speaking. Like it's a total paradigm shift"

·         "[M] Exactly. And they went step by step, you know, first analyse the speech, then identify errors"

·         "[M] Exactly. So you got to clean up the transcript a bit. And then that's when the fun starts. Use ChatGPT to actually like analyse your speech and give you feedback"

·         "[M] They encourage teachers to, like, play with it, change it, adapt it, you know, and even share their own versions with other teachers"

These examples show the more conversational style captured by TurboScribe, including interjections ("wow") and fillers ("like", "you know"). The two transcripts captured by Rev appear to have had some of these conversational elements removed or edited out, presenting a cleaner version of the dialogue.

On the other hand, below is a comparison of how Rev.com with the verbatim option selected and Turboscribe, which has no such option, handles transcribing. It is very revealing:

Rev.com:

Rev heard ‘boat’ and ‘things’ and as it had been asked to create a verbatim transcript it didn’t examine how likely this sequence of words was.

Turboscribe.ai:

Turboscribe, on the other hand, correctly interpreted the learner’s intention because ‘bought’ and ‘thinks’ are more likely in the context. But this desire to predict what is more likely led it to also correct ‘womans’ to ‘women’, ‘say’ to ‘said’ and ‘are’ to ‘was’ and to insert ‘can’ into the phrase ‘you can guess’

Both Turboscribe and Rev interpreted what I heard as ‘ask her’ as ‘asked her’, but I imagine that is because the /t/ is hard to produce and often omitted by native speakers. Why ‘that’ was omitted by Turboscribe in the phrase ‘said no’ must also be down to frequency. About 161,000,000 results without ‘that’ v. About 26,900,000 results with ‘that’ according to Bing. 

There are other important differences between free accounts using Turboscribe.ai and Rev.com:

The most important advantage of Turboscribe is that it can be used from the age of 13 with parents’ permission. The other factor to bear in mind is that Rev.com only works in English on a free account at the moment. The transcripts are automatically synchronised while listening on Rev.com, whereas with Turboscribe.ai the user needs to move the page up to follow them.

But the key reason for not being able to recommend Rev.com on a mobile device is that it is: 

No longer possible to edit or download transcripts using a mobile device

When using a PC, this is not a problem.

Support on Rev.com is excellent and I haven’t needed to contact support on Turboscribe.ai

Your students can get feedback every time they speak

 Voice Recorder

But which one? I’ve got two apps installed with the same name and nearly identical icons. I’ve now discovered that it was installed from the start as it, like my mobile, is from Samsung.

I wanted to record myself giving a talk at a conference and I know that with my free accounts at Turboscribe.ai and Rev.com the limit is 30 minutes, so I tried rehearsing my talk using this ‘Voice Recorder’. It recorded my rehearsal successfully and when asked produced a transcript of a kind, but with nothing like the accuracy of the two other apps mentioned above.

I uploaded the combined voice and text file to ChatGPT and said, “Can you make a better transcription?”

I then listened to the resulting version, which was much better, although it still included one or two points that were not exactly the same as what I had said. I asked ChatGPT to correct these mistakes.

I then made a screen recording on my Android phone of it being read aloud by ‘Vale’

The second things I did was to upload the same voice and text file to NotebookLM. It produced a very flattering account of what I had said rehearsing for my talk at the #APACELT25 conference. It was fairly accurate in its discussion of my ideas and added some additional ideas of its own. 

This is a summary it:

AI-Powered Speech Feedback for Language Learners

1 source

This presentation outlines a workshop using Turboscribe.ai for speech-to-text transcription and ChatGPT for language analysis to provide students with instant feedback on their spoken English. The process involves recording a story, transcribing it using Turboscribe, identifying and correcting errors, and then using a five-step ChatGPT prompt to generate improved transcripts at different language levels. A sample student recording and its analysis are demonstrated, showcasing how the tools enhance language learning. Finally, the presenter shares resources and methods for incorporating these technologies into language teaching.

It is about fifteen minutes long and I have downloaded the audio file and uploaded it to Turboscribe.ai as it transcribes all the 'like' and 'you know' fillers. 

Here is the whole podcast with a transcript on Turboscribe.ai :

I had planned to record myself giving the talk live at APAC 2025, but for one reason or another I simply forgot to clip on the wireless microphone connected to my mobile. I’ll be giving an improved version of this talk at TESOL Spain in Burgos and, I hope improved again, at IATEFL in Edinburgh. Maybe I’ll get around to recording it live!

 

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.

An experiment with a 5-step prompt with 5 types of GenAI

I decided that it must be possible to include all the prompts for a routine sequence of prompts in one rather long prompt.

I wanted to get:

1. A student transcript with the errors marked so they could try to see their errors
2. A corrected version of the transcript
3. An improved version at the students level (A2, in this case)
4. An improved version at the next level (B1, in this case)
5. An improved version at two levels up (B2, in this case)

The secret was to include the five steps in the prompt but to instruct GenAI to wait for the prompt “Next” before moving on to the next step.

I then took a short recording made by a very good A2 student and used Rev.com to get the transcript. The idea is that the student has to copy the long prompt from a WhatsApp group, for example, and paste it into the chosen GenAI and then copy the transcript from Turboscribe.ai or Rev.com  and paste it into the same GenAI.

I did this with these 5 types of GenAI:

1. ChatGPT
2. Gemini
3. Claude
4. Copilot
5. Deepseek

I then copied the output from each GenAI for each step and pasted it into Pearson’s GSE Text Analyzer ( https://www.english.com/gse/teacher-toolkit/user/textanalyzer ), which gave me GSE and CEFR levels for all 25 versions of the transcript.

Apart from looking carefully at the English used in each version to try to understand what language had determined the levels given for them all, I also made an Excel spreadsheet with the two sets of levels. With these I was able to produce two graphs. The first one shows what happened with each type of GenAI:

At first glance Claude was the best at producing increasingly sophisticated versions of the student’s transcript, although they were always half a CEFR level too high. Mark you, the original transcript was already half a level higher as she was a very good student.

Equally obvious is the fact that Copilot was useless!

ChatGPT, Gemini and Deepseek failed to produce increasingly sophisticated versions across the four levels, so they didn’t do what I had intended.

In fact, as can be seen in this second chart, the whole idea didn’t work on average:

Once again, this may be as a result of the student’s original transcript being higher than expected (A2+ rather than A2). Maybe the lesson to be learnt from this is that instead of using fixed levels for each step up, the different GenAIs may be able to produce new version of the transcript one level higher on the CEFR scale. This would have the added advantage that the same long prompt could be useful for all students in a class and across all courses.

Here is the original version of the long prompt derived with help from Turboscribe and ChatGPT’s advice. (I hope the concept of including steps and a trigger word will be useful):

“You will be provided a transcript as well as instructions about what to do with that transcript. Unless otherwise specified, your response should be in the same language as the transcript.[1] Here are your instructions, which you must follow:

Instructions:

1. I will provide a transcript. For each step, you should follow the instructions carefully.
2. For each task, do not include timestamps unless specifically requested.
3. After each step, wait for me to say "next" before proceeding. 

Steps:

1. Mark the errors in the transcript in bold, but do not correct them.
2. Correct the errors, marking the changes in bold and leaving the original errors in brackets.
3. Improve the transcript for A2 level students, marking improvements in bold.
4. Improve the transcript to a B1 level, marking changes in bold.
5. Finally, enhance the transcript to a B2 level, marking improvements in bold. 

[Sample transcript of an A2 student’s recording:]

The history start in the summer of 2011. Anna went on holiday with some friends on island. The photo was taken on hill. Hill is a little mountain and called Ana. The photo is important for her because the stone is a mysterious for her. And she put your hands around it, the stone, and, and she was sleeping. And this photo there are in other places because in your mobile phone, computer, et cetera.”

---

[1] These two sentences came from Turboscribe’s Custom prompt

Warning – very long post! (Remixing) Artifacts with Claude

“Artifacts are content made with Claude like code snippets, text documents, graphics, diagrams, or website designs, that appear in a dedicated window alongside a chat with Claude.

This Artifact was made and published using Claude.ai, a next generation AI assistant made by Anthropic that’s built for work and trained to be safe, accurate, & secure.

You can also remix this Artifact with Claude using your own Claude.ai account to evolve and build on the ideas of the publisher.”

Apart from the fact that Claude can help you code a web app (Artifact) from scratch, anyone using the artifact can also remix it -  not only the creator. So you can do it, too, if you have a free account with Claude!

For example my original vocabulary training artifact didn’t allow the learner to choose to go from English to Catalan or vice versa, but it was just a matter of  asking Claude to add that feature, and it was done. (See the prompts I used below.)

Here is the improved version:

• Claude Artifact             (Animals, Journeys, Verbs – English and Catalan)

I later decided that more people would be able to use the same artifact with translations to and from Spanish rather than Catalan. So, I remixed it. This is the one I shared on social media:

• Claude Artifact             (Animals, Journeys, Verbs – English and Spanish)

Next, I wondered how easy it would be to change the three categories and the lists of words and discovered that I could do that as well. Here are two more versions of the same artifact:

• Claude Artifact             (Travel and Tourism, Health and Fitness, Crime and Safety -  English and Catalan)
• Claude Artifact             (Shopping and Consumerism, Social and Cultural Activities, Education and Learning – English and Catalan)

Here is a list of the prompts I used to remix the original artifact and Claude’s replies in italics:

·    Can you include a toggle between English to Catalan and Catalan to English?
I'll modify the artifact to include a toggle for learning direction between English to Catalan and Catalan to English.

·    Could I change the languages to be Spanish and English?
I'll modify the vocabulary data and labels to use Spanish and English instead of Catalan and English.

·    Can I change the vocabulary lists in this artifact?
Yes, you can modify the vocabulary lists in the artifact!

·    Can you remix the artifact with these three topics and vocabulary lists?
Travel and Tourism tourism, sightseeing, destination, route, adventure, urban, jungle, mountain, trail, city, beach, wetlands, hotel, hostel, tent, exploring, hiking, cycling, running, flights, arrival, connection, Christ the Redeemer, museum, landmark, river, Amazon, Pantanal
Health and Fitness cold, flu, virus, bacteria, infection, handwashing, soap, warm water, sneezing, fitness, gym, yoga, spinning, Pilates, swimming, tennis, basketball, fruit, vegetables, fish, healthy, fried food, fizzy drink, brain, puzzle, reading, learning, exercise, sleep, oxygen, muscle
Crime and Safety robbery, theft, burglary, stealing, thief, burglar, pickpocket, grab, faint, disappear, escape, report, police, investigation, suspect, connection, safety, lock, stranger, attention, train, theatre, bridge, house, glasses, hat, blonde, brown
I'll modify the vocabulary Data object with the new vocabulary lists you've provided. I'll translate them to Catalan to maintain the bilingual learning approach of the original artifact.

·    Can you make a new version with this vocabulary? I'd like to be able to send links to the two versions Shopping and Consumerism shopping, sale, gift, Secret Santa, delivery, jewellery, furniture, accessories, leather, clothing, vegan, organic, handmade, vintage, buy, sell, borrow, advertise, pay, fast fashion, marketplace
Social and Cultural Activities art, gallery, studio, café, urban farm, artist, family, friends, socialising, meeting, performance, opera, singer, judge, audience, festival, concert, event
Education and Learning school, term, course, schedule, reading, studying, practising, learning, revise, memory, thinking, lesson, assignment, classroom, holiday, start, complete
I'll create a new version of the Vocabulary Trainer with the new vocabulary lists you've provided.

 

Please experiment with the artifacts I have shared here and make changes to the language your students might want, and change the categories and vocabulary lists to suit your classes. You could even try remixing the artifact to increase or decrease the number of categories!

Remember that you have to Publish your new artifacts to be able to share them with your students and they don’t need to be 18 or have an account with Claude.

The first time I remixed the original artifact, Claude offered these alternatives:

I'll help you explore variations of this vocabulary trainer. First, let me reproduce the artifact to ensure we're working with it properly.

…

Here are three interesting ways we could remix this vocabulary trainer:

1. Spaced Repetition System (SRS)

We could enhance the learning algorithm by implementing a proper spaced repetition system. Instead of just reviewing unknown words once, we could:

- Add confidence levels (1-5) instead of just Yes/No

- Track when each word was last reviewed

- Calculate the next review date based on performance

- Show daily review queues

- Add a progress streak system

 

2. Multi-directional Learning Mode

We could add flexibility in how users practice:

- Toggle between English→Catalan and Catalan→English

- Add multiple-choice mode where users select from 4 options

- Include a typing mode where users must type the translation

- Add audio pronunciation (text-to-speech)

- Include example sentences for context

 

3. Gamification and Social Features

We could make it more engaging by adding:

- Points system based on correct answers and speed

- Daily challenges with specific word sets

- Achievement badges for milestones (100 words learned, 7-day streak, etc.)

- Leaderboards for different categories

- Option to create and share custom vocabulary lists

- Study groups where friends can track each other's progress

 

Would you like me to implement any of these variations? I can focus on the specific aspects you find most interesting.

I only took up the - Toggle between English→Catalan and Catalan→English option. You might like to try some of the others!

Just a footnote about a mistake I made. I unpublished an artifact and failed to read the following warning.

Warning:
Once Artifact is unpublished, it cannot be republished.
A new Artifact must be created.