D
Deleted member 375592
yes.In your example with HF speaker correction, its IR is in the rir.mat file?
D
Deleted member 375592
1. Our ear (=brain) is not onmi thus a correction filter for an omni IR does not make the ear IR anechoic. It will sound different, sure. But any claims that it solves room reverberation are false.Mike, could you explain to a poor simple soul like me why Dirac in your view is a scam?
2. As you have seen above, the essential part of it is 20 or 30 lines of Matlab code. All the noise is marketing.
Why quibble over details when our rec/rep system is an unclear whole with a fundamental lack of an architecture 😉
//
//
D
Deleted member 375592
TNT, let's look back to the path from the first Edison devices to today's status. Quite entertaining, isn't it? And so it will remain, IMHO🙂
We can have an architecture if and only if we already understand all parts of the problem in fine details we are solving (eg, program control approach). That's not the case and never has been. The feedback (or/and adaptive) control approaches are the correct approaches in such cases. The road is composed of small steps (with unavoidable mistakes), where each one is an (ideally) honest attempt to improve above the existing state of the art while learning from the mistakes and arguing is a part of it. Only to improve - not to solve the problem once and for all. Step by step, every one of us contributes to the eventual progress, and everyone's contribution is valuable. Ok?
We can have an architecture if and only if we already understand all parts of the problem in fine details we are solving (eg, program control approach). That's not the case and never has been. The feedback (or/and adaptive) control approaches are the correct approaches in such cases. The road is composed of small steps (with unavoidable mistakes), where each one is an (ideally) honest attempt to improve above the existing state of the art while learning from the mistakes and arguing is a part of it. Only to improve - not to solve the problem once and for all. Step by step, every one of us contributes to the eventual progress, and everyone's contribution is valuable. Ok?
Not so OK as see it. One don't experiment away for an architecture. It is thought out. The path we are on now is the one of local optimisation without knowing the final goal. I'm doing it myself of course at times... but when we all do this, we must not forget and sound to sure about things....
//
//
I don't know how to use Matlab and I'm not an expert in digital signal processing. I can barely distinguish a pulse from a step.Then you can control pre-echo vs better inversion.
As I understand the essence of the condition for the presence of phase linearity - if the system has a linear phase, then the pulse must be symmetrical, while if the system has a limitation on the signal transmission frequency, then the pulse will have a wavy tail, the tail which means that the frequency bandwidth is limited, and as I understand it, if we want to have a linear phase in a limited frequency band, then to ensure the condition of a linear phase, the pulse must be symmetrical, i.e. the tail will appear before the beginning of the pulse to achieve symmetry. These are all my inventions, guesses. I am not an expert in this field at all.
Personally, I find FIR filters interesting from the crossover side, since they allow you to make a steeper cutoff without twisting the phase at the cutoff frequency, which allows you to squeeze more out of the speakers, but when using an FIR crossover, the so-called pre-ringing appears, a ringing that is sometimes heard when the speakers are not optimally placed and sized.
Could you synthesize a txt file of phase-linear FIR low-pass and high-pass filter coefficients without pre-ringing, for 512 taps of 48 kHz with a cutoff frequency of 2500 Hz?
I will load these filters into the DSP processor and look for the presence or absence of pre-ringing in each band at a frequency of 2500 Hz.
D
Deleted member 375592
Yriy, an FIR filter does not have to be symmetrical or phase-linear. It can be anything you please. Generally, the level of sidelobe in the time domain is inversely proportional to the level of sidelobes in the freq domain.
To get a smooth wide-range FR, you need to complementary overlap the sub, midrange, and tweeter, then you'll have a delta-function response excluding the very top and bottom. You can transform a low-pass prototype into a high-pass by 1-b(z)/a(z), etc. Then, when you add sub, midrange, and tweeter TFs, you get b(z)/a(z)=1/1. On STFT/Gabor/spectral display, you'll see the reverberating tails of IRs canceling each other. You can do this trick with either IIR or FIR. Each of the driver IR's will have a pre-echo. All of them together - nope. But first, you need to convert the drivers' TFs into something nearly perfect, otherwise, you'll get a total mess. To perfect them, you first need to solve the problem of generalized RIR inversion, which I addressed. The quality of IR inversion will be revealed as pre-echo, and I have shown how to put it below -100dB. Ok?
PS> "I can barely distinguish a pulse from a step." 🙂 You are kidding me. Ukraine used to make space rockets, the education was top-notch. You surely have an MSc. Have you ever been to https://dsp-book.narod.ru/books.html?
To get a smooth wide-range FR, you need to complementary overlap the sub, midrange, and tweeter, then you'll have a delta-function response excluding the very top and bottom. You can transform a low-pass prototype into a high-pass by 1-b(z)/a(z), etc. Then, when you add sub, midrange, and tweeter TFs, you get b(z)/a(z)=1/1. On STFT/Gabor/spectral display, you'll see the reverberating tails of IRs canceling each other. You can do this trick with either IIR or FIR. Each of the driver IR's will have a pre-echo. All of them together - nope. But first, you need to convert the drivers' TFs into something nearly perfect, otherwise, you'll get a total mess. To perfect them, you first need to solve the problem of generalized RIR inversion, which I addressed. The quality of IR inversion will be revealed as pre-echo, and I have shown how to put it below -100dB. Ok?
PS> "I can barely distinguish a pulse from a step." 🙂 You are kidding me. Ukraine used to make space rockets, the education was top-notch. You surely have an MSc. Have you ever been to https://dsp-book.narod.ru/books.html?
What do you consider an "omni IR"? One that includes room effects? I am asking because your terminology is not always easy to follow for me.omni IR
D
Deleted member 375592
IR that was taken with an omni mic. Every mic pick-up some room effects, some more (like omni), some less (cardio, super-cardio, figure-8, mic arrays). Our ears are only cardio in the 1st approximation, but our brain is capable of distinguishing 0.75 degrees and is amazing in separating the direct and late sound. When you add reverberation to the room, the human Word Error Rate (WER) increases by only a percent or two, but ASRs (the best ones) have 80% WER. Blind dereverberation improves a bit... but not too much. The demand for such systems is obnoxiously huge - but no one knows how to make them.
I wish I knew the official terminology🙂
I wish I knew the official terminology🙂
Boden, is it understood that there is no substitute for an acoustic solution. If you have control there, including speaker design, you may achieve the ideal?
Please decipher the abbreviation - TFs, I can’t find its decipherment and therefore I can’t understand the main meaning of your message.But first, you need to convert the drivers' TFs into something nearly perfect, otherwise, you'll get a total mess.
No, I haven't been there, and I don't plan to go there. Soviet textbooks are oversaturated with dry mathematics, and sometimes and often divorced from reality. I don't have such a high level of mathematics that I can find answers to my questions in textbooks.Have you ever been to https://dsp-book.narod.ru/books.html?
By the way, I had one teacher who said that all textbooks in the Soviet Union were written for spies, they were written in such a way that without the help of a knowledgeable teacher, it was impossible to understand them on your own. This teacher, although he was joking, but there is a lot of truth in his joke. The textbooks were written by scientists for scientists, and not for students.
Signal processing is not my specialization, although I had to pass a test on the Communication Theory of Secrecy Systems of Mr. Claude Shannon, and this is clearly not enough to delve into the subject of digital signal processing.
D
Deleted member 375592
TF stands for Transfer Function.
I absolutely share your disgust for all things Soviet and I am on your side... but let's avoid politics because nobody in The West would understand our grievances. This DSP library does not contain any Soviet textbooks but lots of texts in English.
I absolutely share your disgust for all things Soviet and I am on your side... but let's avoid politics because nobody in The West would understand our grievances. This DSP library does not contain any Soviet textbooks but lots of texts in English.
Here you say that IR inversion using FIR leads to the appearance of a pre-echo, and you show how to suppress this pre-echo to a level of -100 dB. Do I understand you correctly?To perfect them, you first need to solve the problem of generalized RIR inversion, which I addressed. The quality of IR inversion will be revealed as pre-echo, and I have shown how to put it below -100dB. Ok?
D
Deleted member 375592
No.
I do not know how to explain it any better than I already did. If you don't know DSP math - sorry, but I am of no help.
I do not know how to explain it any better than I already did. If you don't know DSP math - sorry, but I am of no help.
Understood.
Thank you for your time.If you don't know DSP math - sorry, but I am of no help.
I still don't understand exactly what problem you solved, but I understood that your solution does not solve the problem of pre-echo.
- Home
- Loudspeakers
- Multi-Way
- Digital Signal Processing - How it affects phase and time domain