I've just completed a brand new resource called Reading aloud in Spanish, which I have designed to support all students, but especially those studying for GCSE, in reading aloud with accurate pronunciation.
Each page of the resource follows the same format:
- A QR code provides a model of the sound ('ll' in this case)
- In the Prepárate section, students can listen via a second QR code to the pronunciation of six sample words containing the focus sound.
- The final step - ¡Lee! - is the group of phrases and sentences, all containing the focus sound, the sounds from previous pages, and those sounds which are more or less the same in Spanish and in English (C (/k/), CH, F, L, M, N, P, S, T, X), for students to practise reading aloud.
- The sounds included are those which make Spanish significantly different to English - a, o, ll, ñ, e, i, u, ce/ci/z, d/-d, qu, b/v, ge/gi/j, r/rr. These are gradually built up, so that students are only ever reading aloud the focus sound, sounds from previous pages and the ten sounds which are more or less like English.
When I was writing this resource, I use some of the notes and word lists that I had from my Spanish phonics resource, but also incorporated as many words as possible from the Edexcel GCSE prescribed vocabulary list. I use AI as a way of making the composition of phrases and sentences quicker and easier.
I alternated between ChatGPT and Claude AI (thanks to Jérôme Nogues for the tip!) for the sentence generation, with a little bit of Google Gemini towards the end. I found Claude AI the most reliable. ChatGPT tended to go through my alphabetical vocabulary list and use each word in turn to create a sentence, meaning that all the sentences started with A and many did not make sense. Claude AI generated more varied sentences and it was easier to adjust the prompts. Generally speaking, I didn't use many of the AI-generated sentences, but rather used them as a starting point for the sentences that I included. Even so, it was much quicker than writing all the sentences from scratch myself.
In the past when creating resources which include sound, I have recorded the audio myself and have had to edit the sound using Audacity, and insert them into the resource. This time I used the text-to-speech generator TTSMaker, which I heard about from Jérôme at Language World last month. I found a couple of the European Spanish voices which gave accurate results, and then linked them to the resource using QR codes - I used QR code Monkey.
The only difficulty I had with audio was getting the AI to say sounds rather than letters of the alphabet, such as /x/ instead of jota, so I used recordings I already had for those sounds.
I first experimented with AI in August 2023, and wasn't very impressed. I can see that the AI generators have developed and moved on since that time, and now give better results. I'll try again to use it to develop new resources for the classroom, and also to create more audio for the LiPS Listening Project.