Newbie question about bitrate

[yetanotherlogin]

Hello

I downloaded the video of a conference which has been extremely compressed, both video and audio. Bad video is OK but listening to super-compressed audio gets tiring fast It sounds like bad, "bubbly" MP3.

Anyway, out of curiosity, is it possible to compute the sample size given those infos:

Codec: AAC
Channels: Stereo
Variable bitrate:
Bitrate: 16000 bps / 128 kbps
Frequency: 22050Hz

Is it 128000 / 22050 = 6 bits per each sample?

Thank you.

[pandy]

bitrate is variable - 16 - 128 kbps, 2 channels, 22050Hz sampling so bit per sample vary from 16000/(2*22050) to 128000/(2*22050) thus 0.362 - 2.902 bit per sample

[yetanotherlogin]

Thanks much for the info. 2.902 bit per sample would explain why sound is so bad.

[pandy]

Thanks much for the info. 2.902 bit per sample would explain why sound is so bad.

in fact bits per sample are somewhere between 0.36 to 2.9 bit per sample - 2.9 is quite high (lower compression means higher quality) - important is average bitrate and average amount of bits per sample.

[yetanotherlogin]

Good to know. Since a good original recording is 16 bits per sample * 48Hz rate, I thought 3 bits per sample would be quite low.

So what could be the reason why the audio track sounds so terrible? Here's a sample.

[pandy]

Good to know. Since a good original recording is 16 bits per sample * 48Hz rate, I thought 3 bits per sample would be quite low.

So what could be the reason why the audio track sounds so terrible? Here's a sample.

There can be many reasons - probably overall low bitrate like average 32 - 48kbps for example, maybe poor encoder and/or low quality encoder settings, perhaps incorrectly prepared material to encode (like full bandwidth instead lowpass/bandpass filter to focus on most important part).

[jman98]

Good to know. Since a good original recording is 16 bits per sample * 48Hz rate, I thought 3 bits per sample would be quite low.

So what could be the reason why the audio track sounds so terrible? Here's a sample.

The frequency of 22 kHz is definitely part of the reason, if not all of it. Lots of morons believe that the lower the frequency, the smaller the file, but that's not true. Also, in some cases the person making the file knew that 22 kHz was crap and it didn't result in a smaller file size over using 44.1 or 48, but they were wanting to play the file on an older mobile telephone that couldn't handle anything above 22 kHz. So the person who made this is either stupid or they're trying to support a really old phone for playback.

Finally, based on a few posts we've had, a few people have this erroneous idea that because human hearing tops out around 20 kHz or so that a 22 kHz sample rate is more than sufficient to accurately represent sound. Actually, it's been well known that you have to sample at at least TWICE that rate or it always sounds like crap. There are mathematical reasons for this, but I'm not going to go into that. So this kind of erroneous thinking is also part of why stupid people think that 22 kHz is fine (those people also believe that it leads to smaller files sizes).

[pandy]

Small speakers, portable devices and occasional listening in open air can justify sample rate reduction, also there is no sense to go for 48ksps if the recording is pure voice instead for example music. It is good to reduce amount of information before encoding - for example to keep natural sound of human voice, spectrum between 150Hz and 6 - 7kHz is more than enough - everything else can be (and should be filtered) then saved bits can be dedicated to information. Quite common nowadays habit to listen music without earphones on a smart phone justify half sample rate and even mono channel configuration. Speaker probably will not reproduce anything useful lower than 200Hz and higher than 8 - 9kHz. 40 - 50 years ago people listen radios in same way - FM up to 9 - 10kHz, AM up to 4.5kHz.

[yetanotherlogin]

Thanks. The audio file is actually ripped from an MP4 video that was recorded with a camcorder, not a smartphone, so I assume the person recompressed the audio further but still using the AAC format.

Just for my information: Do you recommend recording conferences (ie. just voice) at 44Hz sampling rate in stereo? I can't select the sample size on my Sony camcorder (hi-def audio is "AC3 (5,1, 48000Hz)", while low-def is "audio A52 (AC3, a52) 48KHz").

[pandy]

Thanks. The audio file is actually ripped from an MP4 video that was recorded with a camcorder, not a smartphone, so I assume the person recompressed the audio further but still using the AAC format.

Perhaps it was recorded already in AAC but encoder can be simplified (reduce cost, power consumption) thus poor quality.

Just for my information: Do you recommend recording conferences (ie. just voice) at 44Hz sampling rate in stereo? I can't select the sample size on my Sony camcorder (hi-def audio is "AC3 (5,1, 48000Hz)", while low-def is "audio A52 (AC3, a52) 48KHz").

Low def should be OK (seems that difference is only in channel configuration - 5.1 vs 2.0 - AC-3 is used for both settings).

[yetanotherlogin]

For some reason, the "REC mode" and "Frame Rate" settings on my Sony camcorder are disabled, both in shooting mode and reading mode.

I'll just leave it in STD mode, and compress sound down further on the PC later.

Thank you.

[jman98]

pandy's answer is EXACTLY what I am talking about. The point is not "Is the audio audible at 22 kHa?" Or course it is. The point is "Does the audio sound like crap at 22 kHz?" Yet it does. Do a test. You'll see that I am right. Using 22 kHz introduces artifacts.

Any sample rate at 44.1 kHz or above will be fine for giving you listenable quality. Speech only can usually go down as low as 96 kbps and sound OK although lower bit rates may introduce artifacts.

[pandy]

pandy's answer is EXACTLY what I am talking about. The point is not "Is the audio audible at 22 kHa?" Or course it is. The point is "Does the audio sound like crap at 22 kHz?" Yet it does. Do a test. You'll see that I am right. Using 22 kHz introduces artifacts.

I've started writing this many times and still can't find proper words to reply then i will use simple words - sorry for that - You are wrong and misunderstood what i've wrote previously.

First You are confused with sampling rate and bandwidth, second you are completely ignoring render/target device, third you completely ignoring fact that compressor(codec/encoder) need to use some bits even if there is no information (usable information) in particular part of spectrum and this produce less efficient compression (and thus worse quality).

Anything sampled at 22050Hz will have usable bandwidth up to 11025Hz - this is general rule - with proper preprocessing (antialias filtering) there is no distortions/artifacts introduced by sampling (please simplify discussion and skip quantization errors etc). If our source have no useful information (for example this is lector voice) then there is no sens to push sampling rate up over 22050Hz.

Any sample rate at 44.1 kHz or above will be fine for giving you listenable quality. Speech only can usually go down as low as 96 kbps and sound OK although lower bit rates may introduce artifacts.

Most of human voice energy is located in spectrum area 300 - 3000Hz however such sound is not natural, spectrum are 150 - 9000Hz provide normal, naturally perceived human voice and if there is no music there is no sense to over this. So even if we keep higher sampling rate (like 44.1 or 48ksps) it is recommended to lowpass (or even better bandpass) filter such source before encoding, cutting off higher than 9000Hz improve overall compressibility, cutting off low frequencies eliminate lot of distortions from air flow, power network, poor room acoustic which also affect compressibility (sometimes high energy in low area can saturate encoder and produce pumping like modulation for remain part of spectrum).


In case of AC-3 there are only 3 allowed sample rates thus 22050Hz is not a problem for DD.

[Cornucopia]

The person who made this is just stupid when it comes to audio. While it's likely that the original recording from the cam/phone/device was stereo, there is no reason why the final copy should not be in mono. It's a video conference!. If filesize/bitrate were the issue, a 22kHz 128kbps Stereo file sounds twice as bad as a 22kHz 128kbps Mono file of the same filesize. That's because in the stereo file, the 128kbps is split between both channels (unless one uses joint, etc) whereas in the mono file all 128kbps is used. So 128 per channel vs. 64 per channel...

I guarantee you, I can make a 22kHz 128kbps stereo voice signal sound good and a 22kHz 128kbps MONO voice signal sound great, so if it is less than even fair, they did a disservice to the track.
The lower the sample rate, the smaller of the file IS true with LPCM or ADPCM. It of course is NOT true for lossily-compressed files (though it still does come into play).
BTW, your original estimate of bits per sample are WRONG. 16bits is per sample PER CHANNEL. With stereo, that's 32bits per sample.

Sorry to contradict you jman98, but 22kHz is NOT the reason (or at least here, the main reason). For voice, pre-filtered & optimized tracks only have a little of their "openness" removed at 22kHz compared to 44/48. I can provide samples if you'd like.

@OP: I recommend recording EVERYTHING you record at the BEST possible configuration you can afford. Then, make edits, etc. and give COPIES that are downsized to match the intended audience/playbackdevice. Any other way is either lazy or shortsighted. You unfortunately are ALREADY at a disadvantage because you are recording your ORIGINAL as a compressed file in any case. If you are doing this for a business I would STRONGLY suggest you get a separate audio recording device (I've got a Zoom H4N, but you may not need that much).

Scott

[hello_hello]

If we're referring to variable bitrate lossy compression then it's really got be one or the other. As a general rule, the higher the fidelity the higher the bitrate required so if a lower sampling rate drastically reduces the quality then the bitrate required should reduce accordingly. Plus I really don't see how you can sample something at "X" rate, then resample it again at "half of X" rate and not reduce the file size regardless of the output format. Well except for lossy constant bitrate compression.

Out of curiosity, I picked a random flac file from an album on my hard drive and converted it to MP3 and AAC. MP3 using the standard LAME V2 preset and q.50 for the NeroAAC encoder. File sizes were 7.4MB and 6.7MB
Same encodes again, this time resampling it at 22kHz on the way through. File sizes 4.6MB and 2.9MB

How different did they sound? Well to be honest nowhere near as different as I expected. My PC speakers aren't the ultimate in Hi-Fi but they're THX approved so they don't sound too tragic. In fact the difference was so small I actually ran a third pair of encodes at 16kHz just to make sure the resampling filter was working properly. At 16Khz they were definitely lacking in high frequencies and heading towards the "crap" stage but that's for a CD track and not just a vocal. File sizes 3.7MB and 2.0MB. Average bitrates 102Kb/s and 54Kb/s.

To be honest, for what it is, the sample from the video conference doesn't sound as bad as I expected it to. It sounds like a mic built into a phone/camera was used to record a voice which was amplified through a loadspeaker in a boomey room. Sure there's some unpleasantness which sounds like the filtering wasn't adequate for the sample rate, but we don't know if this was a budget camera or a $10,000 model. It'll never be Hi-Fi but EQ'ing out some of the low frequencies and maybe enhancing the high-midrange a bit might make it sound clearer. Audacity has a bunch of filters you could experiment with.
And of course there's any conversion to a fixed bitdepth on the way through to take into consideration. The quantisation noise could be due to crap converters/sampling and/or a low bitdepth and little to do with the sampling frequency itself. Now the audio is in a format without a fixed bitdepth it's anyone's guess.

I did find myself wondering about the stereo/mono theory for a vocal recording given when using a lossy encoder, a dual channel mono recording should use the same bitrate as an identical single channel one. So I converted the sample to stereo wave, then to a mono wave, and then those to MP3. So yes, it appears there's enough stereo there to make a stereo type difference to the file sizes. Same V2 Lame preset again. Stereo 247.1KB, mono 126.2KB (file sizes).