Hi. Please add 64-Bit Float audio processing support in the next Audacity updates. You also need sample rate up to 768khz, 1536khz, 3072khz. You also need the ability to export files with a frequency of up to 3072kHz and 64 Bit Float in PCM (WAV and WavPack) format. You also need to add point sinc resampling up to 1024 times. This method is good when you use it before exporting and it significantly improves the sound.
I don’t speak for the developers…
People often complain about the size of Audacity projects and 64-bit audio data would double the size with no practical-audio benefit.
Floating point makes processing easier, and it allows you go go-over 0dB, but most 24-bit ADCs & DACs are only accurate to about 20-22-bits.
And “we” don’t need crazy sample rates for audio.
This method is good when you use it before exporting and it significantly improves the sound.
I’m guessing you’ve never done a proper-scientific blind ABX test? Blind listening tests can be very humbling…
5 Likes
Firstly, increasing the sample rate gives a more accurate and detailed representation of the sound. Secondly, there are DACs with 64-bit audio processing technologies. There is also a soft player with such functions. Regarding the comparison tests, I hear the difference between 192kHz -768kHz, 1536, 3072. It lies in the fact that as the frequency increases, the sound becomes more detailed, more transparent and saturated.
You can hear all the way up to 1.5 MHz? Impressive! That’s over 10x better than bats manage, and well into medicinal ultrasound frequencies.
2 Likes
1 Like
Dear Steve Austin,
I believe your suggested changes will be included in the version of Audacity scheduled for release on the 12th of Never.
The sound frequency (what we hear) and the sample rate are different technical concepts. The higher the sample rate, the more accurately and in detail we hear the original frequencies recorded in the track in the range from 20 Hz to 20kHz (Hi res up to 100khz). In the case of hi res above 20kHz, there are additional overtones and harmonics in the audible part, which make the sound complete and not clamped. Above the frequency of oversampling, this is solely for the sake of detail, not for the ears.
You are correct in that the sampling rate is different from the frequencies we hear, though related. A normal 44100 Hz sample rate (“CD-quality”) encoded signal can at most contain be 22050 Hz. That is called the Nyquist frequency. If you try to put a 30000 Hz signal into a 44100 Hz sample rate, you’ll get distortion (aliasing).
However, where you’re incorrect is that sampling rate is entirely divorced from the frequencies we hear. A 192 kHz sampling rate just means that it can resolve frequencies up to 96 kHz without distorting. Your 3072 kHz number can represent up to 1536 kHz signals. Again: 10x more than what bats can do.
The “additional detail” you see in your waveforms is just ultrasound frequencies you can’t hear anyway. If you see any of them to begin with:
Microphones used in human recordings are designed to give a response in human audible frequencies, ie roughly 20 Hz to 100 Hz on the low end, and 15000 Hz to 20000 Hz on the high end. Anything higher becomes problematic: The nice and big diaphragm mics we need to capture something bassy well just jiggle internally at high frequencies, instead of converting the jiggle to an electrical signal. Because of this, mics tend to drop off in frequency response dramatically at the higher end, often around the 15 kHz mark. To meaningfully capture ultrasound, you need special, tiny mics, which in turn are very bad at picking up the sound humans can hear. So if you’re listening to anything recorded using a studio microphone, the odds of you finding any meaningful “additional detail” above 20 kHz are staggeringly low.
The same thing is true for instruments: They’re designed to wiggle themselves (and in turn, produce sound) at audible frequencies. To create meaningful ultrasonic wiggles, the materials we make instruments of are much too soft.
It of course also is true for speakers. Even something like the Adam T7V, which uses a ribbon as a tweeter (much more lightweight and thus easier to jiggle than a dynamic tweeter!), only gets up to 25 kHz. – something they explicitly boast about!
Because of these factors, any “additional detail” you might be expecting just don’t materializes in reality. But let’s see what happens if the ultrasonic sound somehow did get recorded and you ended up with a file with ultrasonic content.
The file you have contains audible sound, it might contain desirable ultrasonic “additional detail”, or – more likely – the 20 kHz - 120 kHz ballast used in the fluorescent lamps on the ceiling in the studio.
If you just downsample that high sample rate back down to something you can put on a CD and sell to people, you’ll find that the ultrasonic signal is now well in the audible range again (because of aliasing). If someone streams this file with their perhaps lower-quality DAC (phone, laptop, …), it gets downsampled again, and again causes aliased whine.
And nobody wants a constant whine in their recording. Because of that, most audio equipment, especially professional stuff, comes with a low-pass filter somewhere to reduce ultrasonic noise
At the end of the day, manufacturers of audio equipment need people to buy their stuff, and that’s much harder when a device from half a century ago does the job just as well as a new one. So into a rat race they went, improving numbers and theoretical performance, while providing 0 tangible benefit to their human listeners. Your comment reflects that mindset well:
I and @DVDdoug argue that audio should be for the sake of your ears. There is no point in trying to get a better number when your ears cannot tell the difference! And we have strong reason to believe your ears can’t tell the difference between any of these sample rates. We’ve seen countless people come in with this claim, try a blind ABX test and leave with a newfound appreciation of the power of the placebo effect.
2 Likes
Oversampling does not introduce any distortion in the form of ultrasound and other things. This can be seen in the file spectra. Again, I repeat, it’s not about bandwidth, but about the amount of sampling to accurately measure the original sound. The higher the frequency, the more accurately all the details are heard. This is not only my observation, but also that of many people.
I conducted tests and comparisons and took the same track at 48kHz frequencies., 96, 192, 768, 1536, 3072. The difference between them was in favor of increasing the frequency. I also hear the difference between MP3 320, Flac CD quality, HI RES 192. The tests were blind and with the observation of what is playing at the moment. I noticed the difference everywhere.
You may find this video informative. It is from “Xiph”, the people that created the Flac, Ogg Vorbis and Opus audio formats: 24/192 Music Downloads are Very Silly Indeed
1 Like
If you merely oversample, you’ll just create a whole bunch of redundant data, but no noise. That is correct. If you actually try to do something with that increased bandwidth, that’s where the aliasing might appear on downsampling
I’m afraid that’s exactly what the bandwidth is. If you have a bandwidth of 100 Hz, you need at least a 200 Hz sample rate to accurately represent it. Using a higher sample rate than that (eg 1000 Hz) is just a waste of (digital) space.
The full bandwidth for human hearing is about 20 kHz. Therefore, the sample rates to accurately represent every detail humans can hear need to be about 40 kHz. Which is why 44.1 kHz or 48 kHz, respectively are the standard sample rates.
1 Like
I’ve been reading these articles for a long time, and I know what they’re talking about. All this has a biased interest because the OPUS developers wrote the article.
The Kotelnikov and Nyquist theorem states that it is necessary to have at LEAST 2 times the frequency of the original sound in order to transmit the sound completely. Firstly, the theorem itself has problems and errors, and secondly, there is a statement that x2 is the very MINIMUM of oversampling There is nothing said about the maximum and limit of oversampling. So you can do higher things indefinitely.
This is incorrect. The theorem is mathematically sound and has been repeatedly proven, used, and verified in real-world systems for nearly a century.
True, but beyond a certain point, increasing the sample rate offers no audible benefit, and can even degrade fidelity, as demonstrated in the video.
In this context, we are not talking about video, but about our own experience up to 3072kHz, there are sound improvements and differences. No one has checked what will happen to the sound above 3072kHz yet. Although if we take DSD as an example, there are noticeable improvements up to 45 mhz.
You can, and obviously intend to, argue until the cows come home but IT’S NOT GOING TO HAPPEN
1 Like
I wasn’t going to argue, they just started to prove to me that it was useless. What’s the point of that? Adding this to Audacity is not a problem. And which files a particular software user prefers is a personal matter. I’m not proving that MP3 and FLAC are not high-quality formats. Everyone decides for themselves what to listen to.
Just checking that you are aware that Audacity already supports extreme sample rates in the MHz range.
I doubt that 64-bit sample format will be added any time soon, but that’s down to the development team. I don’t personally see a strong case for 64-bit audio as the dynamic range of 32-bit float is already over 1500dB, though I’m aware that some DAWs do support 64-bit float. Adding 64-bit support would most likely be a huge undertaking as Audacity works internally in 32-bit float - From a technical standpoint, it is very unlikely this will ever happen, though as I said, it’s not my decision.
It supports, but it does not support you to process and export to a sample rate of 3 MHz. 64 Bit Float gives increased accuracy in measuring the dynamic range in many professional studio DAWs and software programs it is used. When rendering and mastering in 64-bit, the sound is much better than in 32-bit. Also, some DACs and soft players support 64-bit audio processing.