Lets say I have a sound file in my project that is 70 kb on the disc. How much RAM will it take up? Also 70 kb?

Images take up way more RAM than disc space with scratch. Like, a small 60 kb image on the disc takes up a whole MB of RAM (1200 kg) when opened in a scratch project. I think this is because a lot more info on the image is loaded for easy access to the color rastor, and also because scratch might be storing the image twice (I know it used to store lists twice). So, do sound files get blown up when loaded on scratch in the RAM like image files do?

I've tried testing myself, but the RAM is fluctuating a lot on my scratch project so I can't get an accurate read when loading different versions with different amounts of sound files in them. (Technically I'm doing this on Turbowarp, but I assume its the same for scratch projects).

Trying to plan for a project. If I have 100 MB of sound files in total, will that get completely blown up to multiple gigbytes and crash on systems without much RAM? Or will it be about 100 MB of sound in RAM when loaded?

RAM usage doesn't directly correlate with size of a sound file.
RAM usage (as far as I know) for sounds varies upon the Bitrate. Higher Bitrate should mean higher RAM usage (I don't know I'm not an audio guy)

dynamicsofscratch wrote:

RAM usage doesn't directly correlate with size of a sound file.
RAM usage (as far as I know) for sounds varies upon the Bitrate. Higher Bitrate should mean higher RAM usage (I don't know I'm not an audio guy)

Thanks, this gives me a starting point to look into!

To test this, I made a blank (stereo, 44100Hz EDIT: apparently TurboWarp says it's 48000Hz?) .mp3 file in Audacity that's over an hour long, 1:01:28 to be exact. It was 14.6 megabytes in size. For comparison I also made an uncompressed .wav version (using a 16-bit encoding), which is a whopping 650 MB.

I then opened TurboWarp (because I don't feel like using vanilla, since that would pretty much either require logging out or using the offline editor which I don't have, as the sound is over 10MB even when compressed), put the .mp3 version of the sound into a blank project and saved it.
I then reloaded the page, opened Firefox's Task Manager (which lets you see the memory usage of tabs) and loaded the project. As it loaded, I saw the memory usage increase all the way up to 4GB (Firefox doesn't let me see a more accurate number), but when the project finished loading it dropped back down to 1GB. It also took around 22 seconds to load.

(I was gonna put more detailed findings here involving Firefox's developer tools but then I realized i did all those observerations in the editor, so they probably took up more RAM. The relevant thing you need to know is that sounds in memory are stored uncompressed in RAM and they seem to take up around twice as much RAM as a normal 16-bit encoding.)

So, for measurement, I resaved that .mp3 file as an uncompressed .wav, in a 32-bit format. It took up 1.3GB of space on my hard drive.
Additionally, I wanted to test what other factors can affect the memory usage. So I made 2 more versions of that .mp3 file. One was converted to mono (from stereo), and for the other I lowered the sample rate down to 11025Hz, a quarter of the usual sample rate, and then I saved 2 blank projects with those 2 sounds.
The mono sound project ended up using around 800 megabytes of RAM, most likely half of the full sound's RAM usage. The quarter sample rate project… still used 1GB of RAM.

So: Sounds seem to take up around 2x as much RAM as how much they would take if you saved them as a regular 44100Hz .wav file in a 16-bit encoding, alongside the original compressed sound's file size. However, while loading, they might take up as much as 8x RAM, and it will also increase loading times. They might also take up more memory in the editor too.
The memory usage can't be decreased by compressing the sound or lowering its sample rate, as Scratch uncompresses and reencodes it when loading the project. However, you can decrease it by making the sound mono instead of stereo, though this will also decrease the quality of it a bit.

CST1229 wrote:

(#4)
To test this, I made a blank (stereo, 44100Hz) .mp3 file that's over an hour long, 1:01:28 to be exact. It was 14.6 megabytes in size. For comparison I also made an uncompressed .wav version (using a 16-bit encoding), which is a whopping 650 MB.

I then opened TurboWarp (because I don't feel like using vanilla, since that would pretty much either require logging out or using the offline editor which I don't have, as the sound is over 10MB even when compressed), put the .mp3 version of the sound into a blank project and saved it.
I then reloaded the page, opened Firefox's Task Manager (which lets you see the memory usage of tabs) and loaded the project. As it loaded, I saw the memory usage increase all the way up to 4GB (Firefox doesn't let me see a more accurate number), but when the project finished loading it dropped back down to 1GB. It also took around 22 seconds to load.

(I was gonna put more detailed findings here involving Firefox's developer tools but then I realized i did all those observerations in the editor, so they probably took up more RAM. The relevant thing you need to know is that sounds in memory are stored uncompressed in RAM and they seem to take up around twice as much RAM as a normal 16-bit encoding.)

So, for measurement, I resaved that .mp3 file as an uncompressed .wav, in a 32-bit format. It took up 1.3GB of space on my hard drive.
Additionally, I wanted to test what other factors can affect the memory usage. So I made 2 more versions of that .mp3 file. One was converted to mono (from stereo), and for the other I lowered the sample rate down to 11025Hz, a quarter of the usual sample rate, and then I saved 2 blank projects with those 2 sounds.
The mono sound project ended up using around 800 megabytes of RAM, most likely half of the full sound's RAM usage. The quarter sample rate project… still used 1GB of RAM.

So: Sounds seem to take up around 2x as much RAM as how much they would take if you saved them as a regular 44100Hz .wav file in a 16-bit encoding, alongside the original compressed sound's file size. However, while loading, they might take up as much as 8x RAM, and it will also increase loading times. They might also take up more memory in the editor too.
The memory usage can't be decreased by compressing the sound or lowering its sample rate, as Scratch uncompresses and reencodes it when loading the project. However, you can decrease it by making the sound mono instead of stereo, though this will also decrease the quality of it a bit.

holy
edit: im gonna also test this stuff out in chrome

CST1229 wrote:

To test this, I made a blank (stereo, 44100Hz) .mp3 file in Audacity that's over an hour long, 1:01:28 to be exact. It was 14.6 megabytes in size. For comparison I also made an uncompressed .wav version (using a 16-bit encoding), which is a whopping 650 MB.

I then opened TurboWarp (because I don't feel like using vanilla, since that would pretty much either require logging out or using the offline editor which I don't have, as the sound is over 10MB even when compressed), put the .mp3 version of the sound into a blank project and saved it.
I then reloaded the page, opened Firefox's Task Manager (which lets you see the memory usage of tabs) and loaded the project. As it loaded, I saw the memory usage increase all the way up to 4GB (Firefox doesn't let me see a more accurate number), but when the project finished loading it dropped back down to 1GB. It also took around 22 seconds to load.

(I was gonna put more detailed findings here involving Firefox's developer tools but then I realized i did all those observerations in the editor, so they probably took up more RAM. The relevant thing you need to know is that sounds in memory are stored uncompressed in RAM and they seem to take up around twice as much RAM as a normal 16-bit encoding.)

So, for measurement, I resaved that .mp3 file as an uncompressed .wav, in a 32-bit format. It took up 1.3GB of space on my hard drive.
Additionally, I wanted to test what other factors can affect the memory usage. So I made 2 more versions of that .mp3 file. One was converted to mono (from stereo), and for the other I lowered the sample rate down to 11025Hz, a quarter of the usual sample rate, and then I saved 2 blank projects with those 2 sounds.
The mono sound project ended up using around 800 megabytes of RAM, most likely half of the full sound's RAM usage. The quarter sample rate project… still used 1GB of RAM.

So: Sounds seem to take up around 2x as much RAM as how much they would take if you saved them as a regular 44100Hz .wav file in a 16-bit encoding, alongside the original compressed sound's file size. However, while loading, they might take up as much as 8x RAM, and it will also increase loading times. They might also take up more memory in the editor too.
The memory usage can't be decreased by compressing the sound or lowering its sample rate, as Scratch uncompresses and reencodes it when loading the project. However, you can decrease it by making the sound mono instead of stereo, though this will also decrease the quality of it a bit.

Thanks for this in-depth testing! The early startup high ram usage and then going back down was what I experienced as well in my testing, which I brushed off as fluctuations and discounted entirely. In my testing I had one project with about 2 Mb of sound, and the other had 0. Both on startup were using ~600 Mb of RAM, and dropped down to ~300 after a minute or so. It might have been more pronounced if I had used a larger sample size of sound to compare to.

The mono vs stereo trick I did not expect. In the editor it shows (what I'm guessing is) the compressed file size, and whenever I convert from stereo to mono the file size goes up ~4x, as well as warning you before you do it that you won't be able to convert it back to stereo, so I assumed stereo would be better for RAM.

PurplePyramid wrote:

(#6)

CST1229 wrote:

To test this, I made a blank (stereo, 44100Hz) .mp3 file in Audacity that's over an hour long, 1:01:28 to be exact. It was 14.6 megabytes in size. For comparison I also made an uncompressed .wav version (using a 16-bit encoding), which is a whopping 650 MB.

I then opened TurboWarp (because I don't feel like using vanilla, since that would pretty much either require logging out or using the offline editor which I don't have, as the sound is over 10MB even when compressed), put the .mp3 version of the sound into a blank project and saved it.
I then reloaded the page, opened Firefox's Task Manager (which lets you see the memory usage of tabs) and loaded the project. As it loaded, I saw the memory usage increase all the way up to 4GB (Firefox doesn't let me see a more accurate number), but when the project finished loading it dropped back down to 1GB. It also took around 22 seconds to load.

(I was gonna put more detailed findings here involving Firefox's developer tools but then I realized i did all those observerations in the editor, so they probably took up more RAM. The relevant thing you need to know is that sounds in memory are stored uncompressed in RAM and they seem to take up around twice as much RAM as a normal 16-bit encoding.)

So, for measurement, I resaved that .mp3 file as an uncompressed .wav, in a 32-bit format. It took up 1.3GB of space on my hard drive.
Additionally, I wanted to test what other factors can affect the memory usage. So I made 2 more versions of that .mp3 file. One was converted to mono (from stereo), and for the other I lowered the sample rate down to 11025Hz, a quarter of the usual sample rate, and then I saved 2 blank projects with those 2 sounds.
The mono sound project ended up using around 800 megabytes of RAM, most likely half of the full sound's RAM usage. The quarter sample rate project… still used 1GB of RAM.

So: Sounds seem to take up around 2x as much RAM as how much they would take if you saved them as a regular 44100Hz .wav file in a 16-bit encoding, alongside the original compressed sound's file size. However, while loading, they might take up as much as 8x RAM, and it will also increase loading times. They might also take up more memory in the editor too.
The memory usage can't be decreased by compressing the sound or lowering its sample rate, as Scratch uncompresses and reencodes it when loading the project. However, you can decrease it by making the sound mono instead of stereo, though this will also decrease the quality of it a bit.

Thanks for this in-depth testing! The early startup high ram usage and then going back down was what I experienced as well in my testing, which I brushed off as fluctuations and discounted entirely. In my testing I had one project with about 2 Mb of sound, and the other had 0. Both on startup were using ~600 Mb of RAM, and dropped down to ~300 after a minute or so. It might have been more pronounced if I had used a larger sample size of sound to compare to.

The mono vs stereo trick I did not expect. In the editor it shows (what I'm guessing is) the compressed file size, and whenever I convert from stereo to mono the file size goes up ~4x, as well as warning you before you do it that you won't be able to convert it back to stereo, so I assumed stereo would be better for RAM.

you can test things yourself by using chrome's inbuilt ram monitor