Dictation + Chat GPT + LLM + LINGQ - Monologues

Hello everyone, I wanted to share with you a learning experience.

I just wanted to share with you a change I’ve made in my language learning approach over the past two months. I’ve been experimenting with a new approach, and it’s been very helpful.

Have you ever felt frustrated by the obstacles you encounter when trying to master a new language? I have! From the endless search for suitable material to the challenge of maintaining interest in study, I’ve faced many of the common problems that come with language learning.

Common problems in language learning:

  1. Finding adapted material:
  2. Ensuring the material is interesting and captures our attention:
  3. The time wasted searching for material.

This is the idea I’ve been applying for the past two months, this is the method I’m using.

  1. Dictate your own monologue in your native language: It’s much quicker than finding adapted material since it comes from you. Even as you read a news article that interests you, you can read it aloud using the words you would use while dictating, making it even easier to incorporate vocabulary and grammatical structures at your own level. You can watch a YouTube video and summarize it aloud after watching. At first, I tried making individual sentences, but I think it’s best to go straight to a monologue, a story that’s about 10 minutes long, between 5 and 10 minutes. And do I move on to another video or do another monologue of my own?

  2. Accurate translation with ChatGPT: In this aspect, it’s important to set clear requirements for translation. Usually, if you don’t specify anything, the translation will be suitable because you’ve already lowered the level during dictation. However, if ChatGPT tries to complicate your message, it’s helpful to indicate that it should maintain the level or even reduce it to ensure a comprehensible translation that is appropriate to your level of language proficiency. The truth is that this is working for me because as soon as I import the lesson into the LINGQ program, I see that there are actually a percentage of words in blue, that is, new words, in the amount that I want. I’m really controlling the percentage of new words, because while I’m dictating, I’m very aware of that.

  3. LLM Pronunciation System: The LLM pronunciation system, short for “Language Model for Speech Synthesis,” represents a significant advancement in speech synthesis through artificial intelligence. Unlike conventional synthetic voices, which can sound mechanical and unnatural, LLM uses a more advanced approach that considers the entire context of the text to generate authentic and fluid pronunciation. In this aspect, I lack words to try to say that this is the key that has changed my way of working and learning, of acquiring the language, because without an appropriate voice, you can have problems, as in the case of Chinese, which they call fossilization. Because you can be reading a lot, know a lot of vocabulary, but you’re really pronouncing it incorrectly.

This system analyzes the text in its entirety, identifying grammatical patterns, intonation, and speech rhythm similarly to how a native speaker would. Additionally, LLM can adapt its pronunciation and intonation according to the content of the text, providing a more natural and immersive listening experience.

Benefits of the method:

  • Access to personalized and adapted material.
  • Increased learning efficiency by using personal resources.
  • Improvement in listening comprehension and pronunciation through advanced technology.
  • I’m increasing the number of audio hours considerably, multiplied by 3 or 4. The number of audio hours has skyrocketed. And I also see that, when I listen to it again, since it’s between 5 and 10 minutes long, it’s very easy to play the audio and listen to it a few times. I also see that after a few days, when I enter the lesson.
  • For me, the key is that before, when I listened to something, a lesson, it was material external to me. It hadn’t come from me. So, after a few days, when I entered the lesson, I had forgotten a lot of things. there wasn’t even a connection . When a lot of time has passed, it’s like reading them again. However, the messages you create, the monologues, especially when you talk about things you’re thinking about or environments you know, they grab onto you with enormous strength, and the words kind of support each other and in a context that’s not imagination, maybe it’s something related to your house or your car. I think this connects with memorization systems that use rooms in the house and things that are very close to you. If you do that, the new words when they are incorporated and for example, you talk about a fire and you have a fireplace, you relate it very well, but you’re thinking about the fireplace in your house and the fire. It’s as if you’re experiencing it. So that the new vocabulary stays much more fixed in your head. This is what I’ve discovered.

In summary, the LLM pronunciation system represents a milestone in speech synthesis, offering a more immersive and effective language learning experience. i am using this one but is plenty of them. see this link as example. natural reader llm - Google Search its a bit expensive but it is what i use by the momment

3 Likes

I like the idea (monologue) and am doing something similar, although I’ve been focusing it more on what’s personally happened to me over the course of the day or a trip. I have also, though, done this with news articles or topics that interest me.

One thing I do though before doing my own monologue is to have chatgpt simplify the article to an A2 level. This usually leaves the sentence structures a lot simpler and simplifies some of the fancier words. Lingq “simplify” does this too, but I’ve been messing around with controlling the level of simplification with ChatGPT. In any event, I find what is produced here to be more than adequate for what I would want to say. Due to the vocabulary it may actually end up at a B1 sort of level anyway, but it seems to ensure attempting to say things as simply as possible without sounding too simple.

After the simplification then I can write and/or try to come up with a small bit of a speech or text.

BTW, it was unclear if you first attempt to try and say/write something in the target language? Or are you going ahead and creating the monologue in your native language, then translating it with ChatGPT? If the latter, you might try to say/write as much as possible on your own without help (unless it would be like 80% that you need help on). I usually do these sort of things in this manner so at least I try to say something, or work around trying to say something that might convey the message. In real life you may have to do this when you can’t quite figure out the correct words, so it’s a good exercise, especially not to get flustered when the right words won’t quite come.

What tools are you using for LLM Pronunciation? The most natural sounding voices to me, so far, that I’ve listened to are the ones via Edge Browser (I think you can also get them through some Azure service. Do you have a different tool?

1 Like

"Thank you for your interest! Just to clarify, the monologue I’ve prepared is in my native language, not the one I’m currently learning.
“Additionally, in response to your inquiry, I’d like to mention that lately, I’ve been advancing my skills by directly reading aloud while dictating news articles in my native language. This approach allows me to take a more active role in generating ideas rather than solely relying on ChatGPT.”
“Why does it always seem like with ChatGPT, the dialogues or texts have such childish names? Sometimes, I find myself having to input a bunch of prompts.”
“Another slightly more advanced technique I’ve been using is to provide ChatGPT with a text, like a news article in my native language, and ask it to summarize it. Once it generates the summary, I read it out loud while paraphrasing and changing words and phrases in my own language in dictation mode. Then, I send it back to ChatGPT to translate directly.” But i Insist LLM is key feture. LLM Voices | NaturalReader Help Center this is the one i use but there are plenty of them
natural reader llm - Google Search check this voice and tell me what you think about. In chinese , my target language is awesome.

1 Like

I call this the “Dustin” method after “Dustin Schermaul” who is a very active user here, and also has a language learning YouTube channel with a lot of practical advice, including this method!
When combined with the concepts of creating “speaking islands” it seems to be very powerful. I have tried it a few times.
I had never considered using it for ‘summaries’ of content. That seems quite genius. I haven’t used chatGPT to simplify, but I will!
My focus this spring is on improving listening comprehension, as I got ahead of myself with reading. I want to be able to understand the words that I can recognize better. I will practice this occasionally to build habits to improve my listening, and to begin to build “speaking islands” on various hobbies and topics.
Good stuff!

2 Likes

Thanks for posting this.

My concern is with you generating your own text and having it translated. In my experience the result will be a translation, which often, idiomatically, is not the way a native speaker would express the same thing.

3 Likes

I don’t know…These things have gotten pretty good. DeepL and ChatGPT (4.0 in particular) have gotten really good. Not perfect I’m sure, but darn close (language dependent of course). One could also upload the translation…either their own, or the ChatGPT/DeepL version to the LingQ writing exchange and hopefully get someone to correct it. Even here though, an “editor” may not fully understand the intent you are trying to convey so they may also give errors (not grammar errors but rather losing the meaning the writer intended).

They could also have a tutor or someone correct, but then again that’s money and may not be all that much better than DeepL or ChatGPT.

One thing is for sure, at the beginning stages, one’s own attempts at producing will most certainly not be idiomatically correct so I’d take my chances on DeepL or ChatGPT when I don’t know the translation. I’m sure the native will understand you in most cases, and they can then also point you to a better way of saying things if you ever get the opportunity…or you may eventually come across the better way to say something in your reading or watching of videos.

1 Like

i got your point about translation, but this is new era with AI. this is not google translator

try to give any gram to chatgpt and will be able to do it close to native.

I like creativity and innovation.
I’m interested in text generated where 95% of the words are known words. I don’t know how to do that.
I assume the quality of AI’s target language speech varies by language. I’m interested in Greek and so far just trust native text.