Chat GPT can recommend convolution filters for ROON given certain parameters, and even generate .wav file convolution filter for ROON import. AMAZING.
Can only be worse than actually measuring the room which is easy
3 Likes
So good that you felt the need to post this 3 times.
1 Like
You’ve posted it in almost every section of the forum, but it is not true. It can hallucinate some random noise, but it can’t do anything useful, especially generate convolution filters that are dependent on a multitude of factors not being provided to it in this case.
1 Like
Without any measurements? How would it do that?
1 Like
The answer is simple, it can not 
.
LLMs like ChatGPT don’t understand any concepts of reality and tend to “hallucinate”. That’s why quoting any AI/LLMs is not welcome here.
Convolution/room correction without measuring is senseless.
2 Likes
Unfortunately, I tried many times to get DIRAC to measure my room and it failed every time. I have tried all sorts of microphones and settings in DIRAC. So as an alternative, I tried generating a convolution filter for use in ROON using Chat GPT. I provided detailed information on my room size, furniture location, and sitting position. My understanding is that it may use a Harmon Curve as a starting point and modify it. I am not an engineer. I’m just experimenting. I will try to figure a way to upload the convolution file in case anyone wants to analyze it. Here is a link to the actual convolutioon file generated by Chat GPT and Roon recognizes it. As far as what is actually executing in ROON, and its impact, I appreciate any feedback. https://drive.google.com/drive/folders/1Hgveb7DQsaHVFi4EernopNcZ0oUZ2wlk?usp=share_link
Not welcome here? Are you the site moderator or a censurer? I’m just a tinkerer here. Just experimenting. I am a hospital administrator and retired Naval Officer with MS in IT and MHA, so please show some respect. I apppreciate any feedback people may have on what I discovered. Here is the actual file if people want to comment further. https://drive.google.com/drive/folders/1Hgveb7DQsaHVFi4EernopNcZ0oUZ2wlk?usp=share_link
It’s in the forum rules:
Generative AI
Remember, this is a space for sharing your ideas. We allow tools and apps to assist with translation, punctuation, improving clarity and sentence structure, etc., but long-form AI-generated content is discouraged. Community is for meeting and forming friendships with other people, not large language models
2 Likes
There are other tools. Though I used Dirac in the past and it worked, too.
Calculating a correction of an irregular room based on these things is mathematically close to impossible.
I don’t doubt that ChatGPT pretended to have calculated something
Doesn’t help because nobody knows your room and what an actual measurement would have provided.
2 Likes
I believe some may be misinterpreting this forum rule regarding use of AI to provide information. What I wrote here is not generated by AI. I am providing information on what I discovered that AI can reportedly do. It is absolutely true that Chat GPT can generate Convolution Files that can be used in ROON. Now, as to the value or validity of the convolution file in improving sound in a room, that is the question. Certainly works differently than DIRAC. My limited understanding is that the convolution file generated raises up the end bass, and kept the middle frequencies relatively stable, and dropped the treble a bit. I am not a sound engineer. (and again, none of what I am saying right here was generated by AI).
I never purported that the convolution file generated by CHAT GPT was “valid” or “accurate” or made any significant difference to the sound perceived by my ears. But it is absolutely true that Chat GPT can generate convolution files based on one’s language inputs, and they work in ROON.
I know. But @maddib wrote (my emphasis):
It’s correct because LLM hallucinations damage the discussion, and that’s why posting them is discouraged.
By the same reason, that is, LLMs hallucinate, have no concept of reality, can’t do math reliably, and can’t be trusted generally, whatever convolution ChatGPT created for you is also highly questionable
2 Likes
Seems quite useless if it isn’t. Seems like you should have added this piece of info from the start.
Oh, it can generate some file that can be loaded in Roon. It’s just not based on anything real and highly unlikely to do anything useful.
1 Like
ChatGPT and other chatbots have access to the algorithms and math to do all this. If you provide ChatGPT the right inputs, it can most certainly generate convolution filters. For example, if you upload an REW .mdat file and give it some parameters, ChatGPT will provide an appropriate file in a downloadable format. Will it be perfect for your room? I’m pretty sure it’ll be suitable or functional at least.
This is a really interesting use-case, and I appreciate that you went to the trouble to experiment with this and share your results/impressions. As an irrelevant footnote, this is exactly the kind of thing these AI engines are excellent at. It is possible to mitigate hallucinations by writing prompts that create the necessary boundaries.
The only negative thing I can say is now I have a rabbit hole to jump down. Oh, and yes–I am an actual human.
Thank you, Phillip, for your intelligent input !
So if one does the measurements (which is the hard part) and manages to get them uploaded, it might use some algorithm, maybe, to generate something, with some parameters. At that point, isn’t it easier to just use REW itself (or any other software for generating filters) to create them exactly the way one wants?
You are welcome here. But as already explained using an AI tool here will very likely end up in many questions and still unsolved problems. So, spending a lot of time for a rather doubtful result.
You will often read answers here that are simply focused on a short and reliable thus direct answer. This is just all about providing help.
Please keep in mind, LLMs are mostly not what they suggest to be from an average user’s perspective. And that is exactly the reason, why the forum rules here are pretty strict.
Much more interesting would be to learn more about your mentioned problems with DIRAC and a proper microphone, you seem to have tried already.
While I can’t comment on any firsthand experimentation with your use of AI, where I have had success utilizing AI in audio is with building Muse presets for my reference headphones. The presets are artist-specific and only A/B tested in headphone monitoring. There was a lot of tinkering but the results, to my ears, were significantly better than the default playback EQ settings.