Encoding Audio Files as Lossless PCM (.wav)

[Ronaldinho]

I have a few audio files that I've downloaded (legally) and some are MP3 format and others in WAV. I want to edit the songs just a little bit, and I'm using Sony Vegas to do it. I know Sony Vegas isn't really ideal for editing music, but it's good enough for what I want to do.

After I've finished editing the songs, I want to encode them losslessly with the PCM codec (I believe this is a lossless codec) but I just want to make sure that I'm choosing the right settings...

I don't care about file size at all, and I know that If I convert an MP3 to WAV it is not going to increase the quality, I just want to keep the same level of quality. But I need someone to confirm that I'm encoding the songs properly.

MediaInfo tells me that the WAV files I've downloaded are 16-bit, 2-channel, PCM, 44.1KHz, and this is how I've been encoding my edited files with Sony Vegas. But if I can encode a song like that and as a result have no loss of quality, then why is there an option to choose 24-bit? I mean, I would assume that choosing 24-bit would result in better quality... but at the same time I am confused, because how can 24-bit PCM be any better than 16-bit PCM when they are both supposed to be lossless?

Can someone please explain this to me?

Any help would be much appreciated,

Thanks

[yoda313]

I can't explain the 24 bit but personally I would leave it as 16 bit. If the source files are 16 bit there is nothing to be gained by "upscaling" to 24bit. You can't magically make it better.

[Ronaldinho]

Hmm.. Okay. Thanks for your reply.

I guess I'm just trying to understand how these lossless codecs work. You're probably right though. Sticking with 16-bit is probably best.

[davexnet]

When the mp3 is edited in this kind of editor, it's essentially uncompressed.
If you save it as PCM/Wav, you're maintaining the sound as it is in the editor;
the sound samples are saved as-is, no further lossy compression takes place.

[Ronaldinho]

When the mp3 is edited in this kind of editor, it's essentially uncompressed.
If you save it as PCM/Wav, you're maintaining the sound as it is in the editor;
the sound samples are saved as-is, no further lossless compression takes place.

Okay, I see. That's good to know.

Thanks very much

[poisondeathray]

MediaInfo tells me that the WAV files I've downloaded are 16-bit, 2-channel, PCM, 44.1KHz, and this is how I've been encoding my edited files with Sony Vegas. But if I can encode a song like that and as a result have no loss of quality, then why is there an option to choose 24-bit? I mean, I would assume that choosing 24-bit would result in better quality... but at the same time I am confused, because how can 24-bit PCM be any better than 16-bit PCM when they are both supposed to be lossless?

You lose quality when changing bitdepth (going up , and definitely down), as most programs will dither the conversion . So technically not lossless. You won' t notice the difference going 16=>24 , but it's pointless . It's usually better to stay the same (if you started with 16bit, stay with 16bit) , also it will be smaller in filesize

[Ronaldinho]

MediaInfo tells me that the WAV files I've downloaded are 16-bit, 2-channel, PCM, 44.1KHz, and this is how I've been encoding my edited files with Sony Vegas. But if I can encode a song like that and as a result have no loss of quality, then why is there an option to choose 24-bit? I mean, I would assume that choosing 24-bit would result in better quality... but at the same time I am confused, because how can 24-bit PCM be any better than 16-bit PCM when they are both supposed to be lossless?

You lose quality when changing bitdepth (going up , and definitely down), as most programs will dither the conversion . So technically not lossless. You won' t notice the difference going 16=>24 , but it's pointless . It's usually better to stay the same (if you started with 16bit, stay with 16bit) , also it will be smaller in filesize

Nice explanation, now I understand it a bit better.

Thanks

[Sartori]

I read an electronic music magazine about (music) mastering in 24bit and they have to tweak some of the frequencies when they convert it to 16bit as the conversion affects them (if you`re tempted/intrigued by that , just do a google for it) , but otherwise I`d concur with PDR ie dont convert .

[budwzr]

Bit depth refers to sample rate or resolution. As a general rule, increasing the sample rate on something that's already been sampled at a lower rate is pointless, as PDR mentions.

However, if your editing includes anything that might benefit from a higher resolution, that would be an exception to the general rule.

UpRezzing is usually done to better match a target playback device like how DVD players UpRez SD to HD. In music, that would translate to amplification. If you go 16 to 24 you can play it louder before it falls apart.

[turk690]

If you go 16 to 24 you can play it louder before it falls apart.

Two audio files of the same length with the same content both sampled at the same frequency but one is 16bit and the other 24 will both have the same maximum loudness. Most logic circuitry that is in the DACs of the players that will ultimately output the analogue signal have V+ of either +3.3V or +5V. If it's, say, +5V, the maximum signal swing is about ±2.5Vpeak; one extreme of the 16bit data stream, when all of the bits are 1111111111111111, will produce +2.5V exactly as it will when 24 bits at 111111111111111111111111 is coursed through it.
The only difference will be, there will be more gradations between discrete levels in a 24bit stream than its 16bit counterpart. 24bits will afford more dynamic range (up to 144dB, natch!!) and an even lower noise floor. More bits mean more complex DACs and attendant circuitry, as well as increased storage space requirements. This is good for archiving important original analogue recordings, but whether the 24bit stream sounds better than the 16bit to an untrained ear is another topic.

[budwzr]

Ummm........and the bottom part goes on to say....not resolution of existing 16 bit ...blah blah...and we already agreed on that.

[turk690]

Ummm........and the bottom part goes on to say....not resolution of existing blah blah...and we already agreed on that.

You took that illustration wrongly, which it in itself is wrongly drawn. A more accurate one would have illustrated the two waveforms at the same amplitude (although how that can be drawn without one covering the other, I don't know). The 24bit waveform will simply have more points in between crest and trough of the waveform than the 16bit, indicating the there are simply more levels to choose from (and presumably a more accurate recapturing of the original analogue waveform). Look at the drawing label: "a wider (more choices) range of loudness possibilities", not more loudness per se.
OK I get it, this drawing is a simplistic way of showing how there are many more discrete levels/points between trough and crest of a waveform sampled at with 24bits at your disposal than a 16bit. It would probable be much harder to explain that had the two waveforms been drawn at the same size. But there has to be a better way of drawing this.
But I maintain a 16bit and 24bit audio clip, at their maximum quantization each, will produce the same analogue level, after passing through a DAC that is able to handle both types.
This is similar to 8bit and 10bit video: the latter is more accurate in that it will produce more gray levels, but identical video content of both bit quantizations will produce the same voltage levels (and therefore the same contrast and brightness perceptions).

[budwzr]

I hear what you're saying. The sample frequency determines the resolution? Not the sample size, right?

So bitrate is "sample dynamic range"?

And frequency is "sample rate"?

Which one is spatial, and which one is temporal?

[turk690]

Sampling frequency and quantization (sample size, bit width, etc) are independent of one another but both taken together collectively determine resolution. It's not impossible to have a sampling rate of, say 96KHz, but use it with just 8bits of quantization; interesting artifacts we may or may not hear there. Or the other way around, what about a sampling rate of 32KHz with 24bits?? We can consider as middle ground the most common, which is 16bit with either 44.1KHz or 48KHz. Going more will improve accuracy, and for future format anticipations; not necessarily be perceived as better-sounding. Going a bit less will not hurt that much (in most DV camcorders, a choice is given for 12bits, 32KHz) but can decrease space requirements considerably.

[budwzr]

My loudness comment was theoretical, not considering any DAC.

I once hired a DJ for a party and I gave him a bunch of mp3's converted to Redbook audio, and as he brought the volume up, they got shredded and became pure noise. That's how I came to my opinion.

I should have learned by now not to get into these audio threads

[VidProQuo]

Just a word of warning, Budwzr has demonstrated his ignorance on many, many Internet audio and video forums for years now and should never be believed. This thread is an example, his "understanding" of bit depth and sample rate are clearly wrong. His DJ story is pricelessly ignorant. Turk690 has the correct answers in this particular thread.

[Ronaldinho]

Just a word of warning, Budwzr has demonstrated his ignorance on many, many Internet audio and video forums for years now and should never be believed. This thread is an example, his "understanding" of bit depth and sample rate are clearly wrong. His DJ story is pricelessly ignorant. Turk690 has the correct answers in this particular thread.

Okay, thanks

[pandy]

But if I can encode a song like that and as a result have no loss of quality, then why is there an option to choose 24-bit? I mean, I would assume that choosing 24-bit would result in better quality... but at the same time I am confused, because how can 24-bit PCM be any better than 16-bit PCM when they are both supposed to be lossless?

Can someone please explain this to me?

This one is very simple - any processing (editing) that affect signal level (also clean cut when cut is performed with soft envelope - slopes) will produce new values - most of the audio editors works native or in INT 24bit or higher formats to provide highest possible quality (so quite common is 32 bit float or even 64 bit INT/float or even higher).

Until you keep internal format everything is lossless - problem starts when you trying to save for example 16 bit audio - 24 bit or higher resolution audio need to be re-quantized to 16 bit - this can be made in many ways - or by simple truncation (wrong) or by more refine processing that will reduce quantization errors.

So normal signal flow in audio editor is 16 bit to 24 bit (or higher) conversion - edition - saving as 16 bit (if edition was without level modification) OR re-quantization to 16 bit (from higher resolution).

Few additional notes:
Higher sampling rate is always better (so called oversampling), higher bit depth is always better (more quantization steps).
Special cases for signal processing are for example 1 bit converters with very high oversampling ratio (MASH, Delta Sigma, PWM etc).
With special processing 16 bit audio can offer perceived audio dynamics close to 24 bit systems (dithering + noiseshaping).
Remember - ears can hear audio bellow noise level (noise floor) - common misconception for digital systems is that signal can't be lower than noise level of AD/DA converters - this is not true.

[pandy]

I hear what you're saying. The sample frequency determines the resolution? Not the sample size, right?

So bitrate is "sample dynamic range"?

And frequency is "sample rate"?

Which one is spatial, and which one is temporal?

Strange question - bit depth means is how many quantization steps is used to cover range - ie 1V can be divided by 100 quantization steps or by 1000 steps, or by 10000 steps - more steps more accurate quantization process and this can be named "spatial", sample rate is how many quantization steps is done in 1 second, same principle as for bit depth, more means better and yes, this can be named temporal (time) resolution.

Yes, sample rate (or rather sample period) can be linear (like in audio systems) or can be random (stochastic sampling), or can be special case (sampling with equivalent time - sub Nyquist).

Quantization steps can be linear (like on CD audio) or can be non linear (for example A-Law, u_Law and similar) - non linear means that distance between quantization steps on spatial plane for signal level can be different (for example each step have distance twice of the previous one).

Sampling is LOSSY operation and this is why more samples in time (higher sample rate) and higher bit depth (more quantization steps) is almost always better.

[budwzr]

Just a word of warning, Budwzr has demonstrated his ignorance on many, many Internet audio and video forums for years now and should never be believed. This thread is an example, his "understanding" of bit depth and sample rate are clearly wrong. His DJ story is pricelessly ignorant. Turk690 has the correct answers in this particular thread.

Hahaha, at least I have the balls to keep my same screen name for years. I must have busted yours somewhere, but I don't remember you. I'm guessing JC from Albuquerque?

[budwzr]

Strange question -

No, I'm trying to relate audio to video. Audio is really a quite strange animal to me.

When you change the playback speed of video, the rate of motion changes. But when you change the speed of audio, beat changes and also the PITCH changes. That's very odd to me.

The "DJ Story" is true, and I've never been able to figure it out. Apparently when the MP3's were decompressed to PCM something went wrong to cause extreme distortion when played through the DJ console. I would like to understand that.

It's not that weird that someone can edit video and audio quite well, yet lack the technical knowledge. I find myself using anecdotal evidence to try to understand what's happening. It is not my intent to purposely put out false information, but I have a "non-standard" brain and thinking process.

I will always admit being wrong, as in this case, and will learn from it, as will others. I'll be your huckleberry for the better good, hahaha.

[turk690]

Apparently when the MP3's were decompressed to PCM something went wrong to cause extreme distortion when played through the DJ console. I would like to understand that.

One of the things that change when an original LPCM audio file is compressed to *.MP3, on decompression back to LPCM, is the noise floor rises. This should not be a concern because one of the algorithms in compressing for *.MP3 involve tossing out parts of the file that are masked by other frequencies going at the same time (and therefore not heard anyway). On reconstruction, the missing parts are replaced by random values (noise, which is not heard, as it is largely masked by the bits that are left used for reconstructing). The problem creeps in when an *.MP3 file is compressed and uncompressed several times; the noise floor creeps up with each iteration, to the point where it can be heard (it can only be masked so much), and/or it starts interacting with the existing reconstructed audible bits in weird and grotty ways.
One weapon in the arsenals DJs have is the limiter, which can come in a variety of attack times, hard, soft, parametric, etc. DJs love their limiters because when they are handed material they have little idea about & no time to scrutinize, they just push up the levels on the channel where it is fed (with the idea it is going to be, say, hard-limited so will likely go nicely with the rest of the mixes). When an *.MP3 file that has probably been compressed and decompressed a thousand times is handed to a DJ, the limiter will clip off the tops of the waveform as it's expected to; the file's noise floor by this time, which is probably just a few dB underneath those peaks, will come brightly and gratingly through.
So give DJs original *.WAV files. Or at least *.MP3s that are known & certified to be just one compression away from the original, at 256kb/s or better. (This immediately rules out stuff downloaded from YouTube....)

[budwzr]

I remember in the old days the record companies would "seed" Napster with songs that played OK for 10-30 seconds, then start screeching noise. So you'd end up with a bunch of these duds in your collection unless you listened to the whole song.

Now that I think about it, I'll bet that could have happened too. Hahaha. So that troll is now debunked.

[pandy]

No, I'm trying to relate audio to video. Audio is really a quite strange animal to me.

When you change the playback speed of video, the rate of motion changes. But when you change the speed of audio, beat changes and also the PITCH changes. That's very odd to me.

However video is sampled on 2D spatial plane and time plane (i.e. 3D), Audio is sampled only on 1D spatial plane and time plane (2D in total) - when you change frame rate of video you have exactly same effect as on audio beat and pitch changes - different is only way how you perceived change (different hardware constrains).