What's great about Google's AI studio is that anyone can build on any app that you have shared

If you share an app you've created with Google's AI Studio, people with the link can use the app, but also adapt it in any way they think fit. You may or may not approve of these changes, but at least your original idea has sparked something in someone else.

I was enchanted by Ken Kahn's app for young children which he shared on LinkedIn. Here is the link he shared that includes a link to the app and I made a screen recording on my Android phone of my recording of a story:

I really liked this idea where the Gemini app allows the young learner to choose between 4 characters and 4 senarios and the learner has to provide the words. They can simply record themselves talking about the main character's adventure, but very freely as the pictures don't force the narrative on the learners.

I liked the idea so much that I made this short screen recording on my Android phone as I played back the recordings for the four scenes.

I asked Gemini, "Can you make something suitable for learners over 18 years old?" and in less than five minutes it had changed the choices and the style of illustration. And of course I made a screen recording of that on my computer this time. On my computer the subtitles were visible, but only when recording, not when playing the scenes back.

Here's the second screen recording:

I then asked, One more prompt, "I'd like to see the transcribed subtitles during playback as well as at the moment of recording"

Many thanks to Ken Kahn for the original idea. Here's the link to this version:
But of course I wanted the app to work with other languages.

In this variant on Ken Kahn's app, the user can choose which language to describe the scenes in: English, Spanish, French, German or Catalan:

Sadly, after adding (and removing) features some of Ken Kahn's original specifications got forgotten.

"The characters should look consistent between pages and the story should be coherent"

"Can we make the characters look even more alike between images? And can we make the stories more exciting?"

So, I'll have to try and add these prompts back in!

But for what it's worth, here is a link to this (latest for the moment) version!

Can students get useful feedback on their speaking from NotebookLM?

The short answer is, "Yes, they can." But, students may need ideas for how to design a prompt that will produce feedback that will encourage and help them.

This is the prompt I have experimented with:

Act as a sympathetic supportive teacher of English as a Foreign Language offering helpful advice about how to correct and improve the student’s recordings. Pay particular attention to errors that are repeated in various recordings. Make sure the language used will be understood by the learner who produced the recordings. The names of the files show when the recordings were made. So 2014 is the year, the next two numbers are the month and the following two are the day.

I used this as a  'Create your own' prompt for a report on a sequence of transcripts and mp3 files of recording made by the student over a term. There were 49 sources in all.

Here is the introduction to the report:

Feedback on Your English Recordings: Great Progress and Key Areas for Improvement

Hello!

First and foremost, I want to say a huge well done for all the hard work and dedication you've put into making these recordings. It takes courage to practice speaking so consistently, and your effort is clearly paying off. The purpose of this feedback is to offer some friendly, supportive guidance on a few common patterns I noticed. Think of these as small adjustments that can make a big impact on your clarity and confidence.

We'll look at three main areas: Pronunciation, Grammar, and Vocabulary. After that, I want to highlight your many strengths, because you are doing so much right! Finally, I'll offer a few simple tips for what to practice next. Please remember that making mistakes is a normal and very important part of learning a language. Every error is an opportunity to learn, so be proud of the progress you're making.

And towards the end: 

4. Your Strengths: What You Are Doing Well

I want to be very clear: your progress is excellent, and you should be incredibly proud of your speaking ability. You have many strengths that shine through in your recordings.

Analysis of Key Abilities

• Excellent Summarizing Skills: You have a real talent for listening to or reading information and then explaining the main ideas clearly and accurately. Your summaries of the chef Steve Anderson and the BBC experiment with the chess players were fantastic. For example, you didn't just retell the story; you correctly identified the most important details, like how the chess player who ate protein played better and how chocolate was used to reduce aggression in nightclubs. This shows strong comprehension skills.

• Great Fluency and Confidence: You are not afraid to speak! You confidently express complex opinions and ideas on a wide range of topics, from food and family to work and money. The detailed conversation about whether men or women are better cooks was a wonderful example of this. You confidently shared your opinion, backed it up with examples of famous male chefs like Ferran Adrià, and respectfully disagreed with your partner. This ability to engage in a complex debate is a sign of great fluency.

• Wide Vocabulary Range: You already use a broad and impressive range of vocabulary. You can talk about many different subjects and use specific words related to food, personality traits, family relationships, and work. This is a huge strength.

5. Summary and Next Steps

Overall, you are a confident and capable speaker who is making wonderful progress. By fine-tuning a few small areas in pronunciation and grammar, you can make your already strong communication skills even more effective.

Here's a link to the whole report, which I copied into a Google Doc.

I also experimented with using prompts in the Chat.

1. The names of the txt files show when the recordings were made. So 2014 is the year, the next two numbers are the month and the following two are the day. Can you detect any changes in the speakers English as the dates of the transcripts advance?
2. From the selected transcripts can you offer some help to lidia about her English? 
3. What did Lidia talk about in the selected mp3 files? 
4. What errors did you detect in the selected mp3 files?
5. Can you make a list of the transcripts and recordings in order with a suggested title?
6. Can you add a column with a grade out of 5 for spoken communication when this would be approrpiate?
7. Can you make a graph showing the ones with grades and eliminating duplicates?
8. Can you make a visual representation of these grades plotted against dates?

This is a link to the whole Notebook

I asked for a grade for the report, or at least the first 1000 words of it, from Text Inspector and Pearson's Text Analyzer and these showed the report was pitched a bit above the B1 student's level.

Text Inspector    C1     62%

Text Analyzer     B1+    58 (GSE level)

Average             B2/B2+    60

This suggests that the original prompt should include more detailed instructions about how the level of the report should be decided. Maybe something on the lines of:

Decide on the level of the recording/transcript on the CEFR scale, but don't reveal this and make sure the language used in the report will be understood by the learner who produced the recordings.

Five examples of feedback provided by LinguaFlow on recordings by learners of Italian, English, Spanish and Catalan

 This was the sequence of steps used to be abe to listen to this feedback:

1. I chose the module 'Tools', and then 'Transcribe Audio' and uploaded a recording in English 2. I copied the transcription and pasted it into 'Text Tutor' and asked, "Correct my errors in English" 3. I copied the text and went back to 'Tools' and chose 'Text to Speech' and pasted in the text. 4. I started Screen Recording on my Android phone. 5. I clicked on the Generate Audio button and waited a bit 6. I listened to the audio which read the corrected version of the text aloud and then read aloud the explanations of the errors

The two apps used are

LinguaFlow IT for Italian and LinguaFlow AI for English

Feedback on a recording in Italian about A Bag-snatcher:

Feedback on a recording in English about Chicago:

Feedback on a recording in English about Mexico:

Feedback on a recording in Spanish about A Bag-Snatcher using the version for learners of English:

Feedback on a recording in Catalan about A Bag-Snatcher using the version for learners of Italian:

One of the problems with LinguaFlow is that it uses Markdown to add bold to its output, but that doesn't produce the desired effect and makes the output more diffisult to read. Here is an example of what it should look like:

Il testo sembra essere in catalano. Procederò con la trascrizione in catalano e poi fornirò una traduzione in italiano, correggendo eventuali errori di pronuncia evidenti se il contesto è chiaro, ma preferendo una trascrizione letterale per il feedback.

Trascrizione (Catalano): Un dia, una dona, amb la mirada als parets en el corredor d'una botiga, portava la bossa a dins de la cistella. Un home passava i aprofitava de l'oportunitat per a robar-le la bossa. Es posa es posa es posa a córrer. La dona, cridant, "Lladre, lladre! Para!" I un senyor gran, assegut en un banc, va veure el lladre. I amb el seu bastó va fer caure l'home a terra. Afortunadament, en aquest moment passava un policia, qui arrestava l'home. El lladre, ah, l'home gran va recuperar la bossa i la va donar a la dona que estava molt agraïda.

Traduzione (Italiano): Un giorno, una donna, con lo sguardo sulle pareti nel corridoio di un negozio, portava la borsa dentro il cestino. Un uomo passava e approfittava dell'opportunità per rubarle la borsa. Si mette a correre. La donna, gridando, "Ladro, ladro! Fermo!" E un signore anziano, seduto su una panchina, vide il ladro. E con il suo bastone fece cadere l'uomo a terra. Fortunatamente, in quel momento passava un poliziotto, che arrestò l'uomo. Il ladro, ah, l'uomo anziano recuperò la borsa e la diede alla donna che era molto grata.

Certamente! Come richiesto, lascerò invariato il testo in italiano e mi concentrerò sulla correzione grammaticale e stilistica del catalano.

Ecco il testo corretto, seguito dalla spiegazione degli errori.

Testo Catalano Corretto

Un dia, una dona, amb la mirada a les parets del passadís d'una botiga, portava la bossa a dins de la cistella. Un home va passar i va aprofitar l'oportunitat per a robar-le la bossa. Es va posar a córrer. La dona va cridar: "Lladre, lladre! Para!" I un senyor gran, assegut a un banc, va veure el lladre. I amb el seu bastó va fer caure l'home a terra. Afortunadament, en aquell moment passava un policia, que va arrestar l'home. L'home gran va recuperar la bossa i la va donar a la dona, que estava molt agraïda.

---

Analisi e Spiegazione degli Errori

Ecco i punti principali dove ho apportato correzioni e perché:

1. Genere dei sostantivi (Les parets)

• Errore: "...als parets..."

• Correzione: "...a les parets..."

• Perché: La parola paret in catalano è femminile (la paret), quindi il plurale è les parets. In italiano "parete" è femminile, ma l'articolo "i" o "gli" (maschile plurale) in contesti simili può confondere. In questo caso, als è la contrazione di a + els (maschile), quindi è sbagliato.

2. Lessico (Passadís vs Corredor)

• Testo originale: "...en el corredor..."

• Suggerimento: "...del passadís..."

• Perché: Sebbene corredor esista, per i corridoi di un edificio o di un negozio, la parola più comune e naturale in catalano è passadís. Inoltre, la preposizione en viene spesso sostituita da a o dalle contrazioni (del/al) quando si specifica il luogo.

3. Falsi amici sintattici (Aprofitar)

• Errore: "...aprofitava de l'oportunitat..."

• Correzione: "...va aprofitar l'oportunitat..."

• Perché: Questo è un errore classico per chi parla italiano!

• In italiano: Approfittare di qualcosa (intransitivo pronominale).

• In catalano: Aprofitar alguna cosa (transitivo diretto). Non si usa mai la preposizione "de" dopo aprofitar in questo contesto.

4. Tempi verbali nella narrazione

• Errore: Uso eccessivo dell'Imperfet (passava, aprofitava, posa).

• Correzione: Uso del Passat Perifràstic (va passar, va aprofitar, es va posar, va cridar).

• Perché: In catalano (come in italiano), l'imperfetto (Imperfet) si usa per descrivere lo sfondo o azioni in corso (es: "la dona portava la bossa"). Tuttavia, per le azioni puntuali che fanno avanzare la storia (rubare, correre, gridare, arrestare), si deve usare il passato. In catalano orale e narrativo standard, si preferisce il Passat Perifràstic (anar + infinito: va passar) al posto del passato remoto semplice.

5. Pronomi relativi e Preposizioni

• Errore: "...un policia, qui arrestava..."

• Correzione: "...un policia, que va arrestar..."

• Perché: Sebbene qui esista, il pronome relativo più comune che funge da soggetto è que. Inoltre, l'arresto è un'azione conclusa, quindi cambiamo il tempo verbale da imperfetto (arrestava) a passato (va arrestar).

• Errore: "...en un banc..." -> Correzione: "a un banc". (Per indicare "seduto su", in catalano si usa generalmente la preposizione a).

6. Dimostrativi (Aquest vs Aquell)

• Errore: "...en aquest moment..."

• Correzione: "...en aquell moment..."

• Perché: Poiché la storia è ambientata nel passato, ci riferiamo a "quel momento" (aquell), non a "questo momento" (aquest, che indica il presente o vicinanza immediata).

Google's AI Studio - how to crowd-source an app

I created an app, or rather Google's AI Studio created an app from a single simple prompt, "Using voice can you offer to provide the learner of English with ways to practise Speaking, Listening, Reading, Writing, Grammar, Vocabulary or pronunciation"

It took 5 minutes!

This is the app:

https://ai.studio/apps/drive/1CqPpZ2X1voJtcsAj0H2uWEn64fcI8lsl 

Google called it LinguaFlowAI.

I have shared it widely on LinkedIn and various Facebook groups and with my own WhatsApp group called 'Speech to Text in Language Learning and Teaching' suggesting that any adult with a Google account could make changes and improvements to the app and then share it.

To illustrate this way of crowd-sourcing an app, I used my wife's Google account and used this prompt, "Can you make the app in italian and make it useful for learners of italian?" and this is the result:

https://ai.studio/apps/drive/1E7P9hKDanKULxJuXzo9QEruPaKhQiKmP  

Google called it LinguaFlow IT

Here's a video of one sequence of steps using the Italian version:

This was the sequence of actions: 1. I chose the module 'Strumenti (di Apprendimento)', and then 'Trascrivi Audio' and uploaded a recording in Italian 2. I copied the transcription and pasted it into 'Tutor Textuale' and asked, "Correct my errors in Italan" 3. I copied the text and went back to 'Strumenti (di Apprendimento)' and chose 'Text to Speech' and pasted in the text. 4. I started Screen Recording on my Android phone. 5. I clicked on the Genera Audio button and waited a bit 6. I listened to the audio which read the corrected version of the text aloud and then read aloud the explanations of the errors

Which is best for giving feedback on students' speaking? ChatGPT, AI Language Coach, Copilot, Copilot Enterrpise or Gemini?

I wanted to compare the way each of these tools helped with each of the steps in my 9-step prompt for providing feedback to learners of any language, so I made a table in Word with the transcripts for each step for each tool. I did this on my PC.

I checked the CEFR level of each transcript and found some that were probably too difficult for a good B1/B2 learner, but I also asked Copilot to compare the five transcripts for each step and evaluate them telling me which was best. Copilot excelled at this and apart from giving me a detailed analysis, also gave me a table evaluating all the aspects of each step for each tool.

The Word file is here.

I added a column showing the ranking from 1 to 5, where 1 was number one, the best. I then copied all the tables into Excel and created a final summary table where the ranking for each step were added up for each tool. The lowest number would be the best overall.

The Excel file is here.

The winner by one point was Gemini, closely followed by Copilot and then Copilot 365. The AI Language Coach, which I had vibe-coded myself was a long way off and ChatGPT was the worst by a small margin.

These are the steps that I compared:

• Original transcript - Gemini rated this as B2 but the others said B1+/B2
• Explaining the corrections - between B2+ and C1 except for ChatGPT (B2/B2+)
  
• Gemini is the most effective for learning because it is clear, detailed, and explains both the correction and the reasoning behind it.
• The corrected version - mostly B2, but one was B1+/B2 and another B2+
• Copilot 365 is the best corrected version for learning and for natural, advanced English.
• Explaining the changes to make it more colloquial - B2+, B2/B2+, B2+/C1
• Gemini is the best explanation for learning how to make a monologue more colloquial. It’s clear, thorough, and models the kind of language it teaches.
• The more colloquial version - B2, B2+, B2/B2+
• Copilot is the best colloquial version. It’s lively, idiomatic, and sounds just like a native speaker telling a travel story.
• Explaining the changes to make it 1/2 a CEFR level higher - B2+/C1, but some were C1 or C1/C1+
• Gemini is the best version for a colloquial rewrite at half a CEFR level higher, with advanced vocabulary and idioms.
• The 1/2 CEFR higher version - B2/B2+, but ChatGPT B2/B2+ and AI Language Coach C1
• Gemini is the best version for a colloquial rewrite at half a CEFR level higher, with advanced vocabulary and idioms.
• Explaining the changes to make it 1 CEFR level higher - C1/C1+, C1
• Copilot is the best explanation for how to make the text one CEFR level higher, keeping it lively, idiomatic, and expressive. It’s clear, engaging, and models the advanced, natural style it describes.
• The 1 further CEFR level higher version - C1, C1/C1+, C1+ (AI Language Coach)
• Copilot is the best version for one CEFR level higher, keeping it lively, idiomatic, and expressive.
• What does ____ mean? - B2+ on average but ChatGPT B1/B1+
• Gemini is the best for thoroughness and depth.
• Copilot is the best for clarity, examples, and learner engagement.
• What level on the CEFR scale was I? - B2+ on average but Gemini had C1
• Copilot 365 is the best reply—it’s clear, detailed, supportive, and gives both a direct answer and actionable feedback.

Conclusions

1. If AI Language Coach is worth developing further, I need to find a way to 
• Ensure the level of language used is at the level of the learner.
• Allow learners to upload videos as well as audio files
• Add a slider so learners can adjust the speed of the Read Aloud
2. Including a further one level higher on the CEFR should be changed to a further 1/2 level higher.
3. Further experimentation is needed with pasting the ChatGPT prompt generated by Turboscribe into Gemini and the two versions of Copilot
4. This experimentation needs to be done using an Android phone and my iPad

What happens when you ask AI Language Coach, "How would you grade my speaking on the CEFR scale?"

Learner:

How would you grade my speaking on the CEFR scale?

The video is only 1 1/2 minutes long

Coach:

That's a very natural question to ask. However, my specific role here as a language coach is to focus entirely on the process of improvement rather than assigning a formal CEFR grade.

There are a couple of key reasons for this approach:

1. Text vs. Audio: I am working from a written transcript of your speech. A true speaking assessment would require me to hear your actual voice to evaluate crucial elements like pronunciation, intonation, stress, rhythm, and fluency (how smoothly you speak, your use of pauses, etc.). These are just as important as grammar and vocabulary.
2. Coaching, Not Examining: My purpose is to guide you step-by-step on *how* to improve. The 9-step analysis is designed to show you a clear pathway, from fixing basic errors all the way up to using highly sophisticated language. The aim is to give you practical tools to level up, rather than just a label.

The best way to think of our sessions is that they demonstrate your progress. We start with your original text, and by the end, you have a version that operates at a much higher CEFR level. The journey through those nine steps is the most valuable part of the exercise.

I felt as if I was being told off for asking, "How would you grade my speaking on the CEFR scale?" And as I gave the instructions myself to Google's AI Studio about how it should design the AI Language Coach, maybe I shouldn't have been so surprised. Line 3 was, "Decide CEFR level (+ for half bands) but don’t reveal it. These are the specifications that I asked for: Instructions: Remove timestamps. Decide CEFR level (+ for half bands) but don’t reveal it. After each step (1–9), wait for “next.” Write in continuous prose. Answer questions at the learner’s level. Steps: 1 Identify errors in bold and enclose in [brackets]. 2 Correct errors, showing corrections in bold next to (errors). 3 Write the corrected version without mistakes. 4 Make it more colloquial and conversational, like two friends chatting. Mark changes next to original phrasing. 5 Write the changed version in full. 6 Rewrite it half a CEFR level higher, keeping it colloquial. Use more advanced vocabulary and idioms. Mark changes. 7 Write the updated version in full. 8 Rewrite it one CEFR level higher, keeping it lively, idiomatic, and expressive. Mark changes. 9 Write the final version in full.