"Sample Rate" -- extremely important, please read

kayray · Post by **kayray** » February 13th, 2006, 5:04 pm

i have edited this post following Brad's research; please re-read :)

Hey all,

We've recently discovered that recording with the correct "sample rate" is very important. The standard is 44100Hz, though lower rates work, as long as they're not too low. Recordings made at a higher sample rate might not play in all mp3 players.

Chris (tis) has just added a sample rate checker to our validator, which will warn us when we get files recorded at a non-standard rate. But we have to fix them manually, which degrades quality, so it's REALLY IMPORTANT to try to get all readers to make sure that they're recording between 44100hz and 22050hz.

Please, everyone, check your recording software and make sure the sample rate is set to between 44100Hz and 22050hz. Doesn't matter if you don't know what the numbers mean (I don't) but please check.

Now, this is pretty embarrassing but I can't find a FAQ where technical details like bitrate and sample-rate are mentioned. Surely we have one somewhere? Someone point it out to me?

And when the wiki is made, let's make sure we say:

Bitrate MUST be 128kbps
Sample rate MUST be between 44100Hz and 22050Hz
Bit Depth 16

Thanks!
Kara

BradBush · Post by **BradBush** » February 13th, 2006, 9:13 pm

I have downsampled all of my recordings at 22, as this is "standard" voice sample rate (above 22, and really a normal human can't even hear). I didn't realize we had a problem with this. If you are going to resample any of mine, just let me know, and I will save a new version from the original 24bit 44.1 master files. I don't want mine upsampled. Will fix going forward.

Brad

BradBush · Post by **BradBush** » February 13th, 2006, 9:32 pm

Kara, bit depth is just like bit depth in photos. For photos more bits = more color range, for sound more bits = more headroom or sound range soft to loud. 16 bit is more than enough for what we are doing. If you go to 24 bit, you can get more signal with less noise, but then you have to resample to 16 bit when you are making your MP3 which can add back in noise if you don't know what you are doing. So bottom line ... stick with 16 bit.

Also, I did some research, and couldn't come up with any incompatable MP3 players to 22khz. Not that I don't believe you, but more than half the podcasts that I get are 64kbps 22.05khz (which actually sounds better than 64kbs 44.1 which is why I do mine that way). BTW, the tech specs do specify 128kbps (which I believe is a requirement of archive.org), but no where do we specifiy 44.1 sample rate.

Brad

kayray · Post by **kayray** » February 13th, 2006, 9:33 pm

Thanks, Brad, I don't know if 22050 causes problems, I just know that 48000 for certain does and that 44100 is supposedly "safe".

So if you'll re-sample The Juniper Tree it would be very helpful :)

Or, if you want to do some research and find out if 22050 is also "safe" that would be cool, too!

Kara

kri · Post by **kri** » February 13th, 2006, 9:36 pm

Would the bit depth correspond to "Default sample format" which allows a choice for 32-bit float, 24-bit and 16-bit?

BradBush · Post by **BradBush** » February 13th, 2006, 9:43 pm

99.9% sure that 22.05 is safe. Yes, 48 can throw off some players, because their range is up to 44.1. BTW, the 64kbps files that archive.org creates all downsample to 24khz (except for mine and a few others which were already at 22khz - these it leaves the same.)

We may want to make the technical spec read less than or equal to 44.1 as the requirement or just say 22.05, 24 or 44.01 as all should be safe.

Kri, yes default sample format = bit depth. Keep it at 16bit.

BradBush · Post by **BradBush** » February 13th, 2006, 9:45 pm

Kara,

I'm not going to do a new version until we know there is a problem with 22.05. I have so many files out there with this, that it will be a long time to fix them all.

Brad

BradBush · Post by **BradBush** » February 13th, 2006, 9:55 pm

Also, just spot checked a few of the books I have downloaded in 64kbps (zip files).

First chapters of Chip's American Indian Fairy Tales are 22khz also (checked 128kbps file and it was 22khz also - like mine).

Little Princess and Road to Oz also seem to be in 22khz looking at the 64kbps. I loaded chapter 2 - 128kbps of LP, and it is still at 44.1khz, so that is odd. Seems like the older archive.org converter used to resample from 128/44.1 to 64/22.05, but now it goes to 64/24. Interesting enough, if the original is mono 128/44.1, it goes to mono 64/44.1. Hmmm I see no pattern.

But, I still stand that 22.05, 24, and 44.1 should be OK for most MP3 players.

Brad

peastman · Post by **peastman** » February 14th, 2006, 10:31 am

BradBush wrote:I have downsampled all of my recordings at 22, as this is "standard" voice sample rate (above 22, and really a normal human can't even hear).

The highest frequency a recording can accurately reproduce is a bit less than half the sampling rate. So if you record at 22 kHz, you'll lose any sound over about 10 kHz, which is well within the human hearing range. That's why 44.1 kHz is the standard - it's a little more than twice the highest frequency most people can hear.

Peter

kayray · Post by **kayray** » February 14th, 2006, 10:41 am

Golly, thank you for doing all this research!

There's not reason to fix things that aren't broken, of course.

If 22050 works, we'll leave it alone. *whew*

I have 5 Grimms' Tales with funny sample rates...

Four are at 22050, so I'll leave those alone.

One is 32000 -- think that'll break anything?

I will change the first post here so as not to cause major confusion and panic ;-)

BradBush · Post by **BradBush** » February 14th, 2006, 11:07 am

Peter, of course you are right. The effective frequency is half the sample rate. My bust. But, I still think 22khz is fine for voice recordings.

Per http://www.paranormalghost.com/evp_frequency_ranges.htm

The range of frequencies that young, healthy adults can hear is between about 20 Hz to about 14,000 Hz (some up to 20,000 Hz).
We perceive speech mainly in the frequency range of 500 Hz to 3000 Hz.

The frequencies which have the most significance for speech lie between about 100 Hz and 4,000 Hz.

Human voice ranges from 20 Hz to about 14,000 Hz (typically about 300 to 4000 Hz).

Kara, anything equal or under 44.1 should be OK. If a user has a problem, they can use the 64kbps which will be at a lower rate.

Brad

kayray · Post by **kayray** » February 14th, 2006, 11:37 am

Forgot to mention -- archive.org apparently did not derive the file that had a sample rate of 48000...

thistlechick · Post by **thistlechick** » February 14th, 2006, 2:52 pm

I've also seen a couple at 8000Hz and was not able to open them with WinAmp ..... luckily they aren't in the catalog yet.

BradBush · Post by **BradBush** » February 14th, 2006, 3:00 pm

8000 is very low for quality's sake also. We probably want to set a minimum. Don't know what that would be. maybe 22.05

raynr · Post by **raynr** » February 14th, 2006, 3:34 pm

Some are recorded with 24000Hz and that sample rate doesn't work with my telephone. When I'm on the way I always listen with Oggplayer on a Symbian system. I had to reencode them but I thought it was just a problem with my mobile phone (Siemens SX1) and not a general issue.