I can’t say the English experiment with orchestral folk music was as successful and influential as, say, the Russian or Hungarians (or Copland) achieved in the late Romantic era, but there we are.

Moving a vast archive of text posts to another server is one thing, but transferring podcasts is a whole other endeavor altogether. We are aware of the podcasting setup and have been diligently taking steps to fix the audio ‘enclosures’ in order to get it functioning properly again so we can posts many more podcasts in the future. Stay tuned!

I’m not sure exactly how WordPress expects audio to be added – seems to say that it will Just Work as long as you use a full URL for it, or if you’re using the hosted WordPress, and also make it sound like it should just work. Maybe you’re not using a full URL when referring to the audio file in some of the posts? Another link I found on the WP forums is – I don’t think that’s relevant, but I could be wrong.

Thanks for any help you can provide.

Still, it’s comforting to note that, at 256kbps, it comes out very well indeed.

If you use iTunes, you should have an automatic playlist called “Classical Music” which utilises groupings to let you browse all the recordings of a single piece. Try adding the same “grouping” to multiple recordings of the same work, and see what this looks like. If you had all the data, it could be rather handy.

I then opened each of the sound files in an editing program (Sound Studio) and noticed two things: 1) the wave forms for the lower bit-rate files actually cut off at the times I described earlier, and 2) the VBR files no longer produce the warbling tone that I hear when playing in iTunes. I’m not sure where that’s coming from.

Also of interest – the wave forms in the ‘pure’ 128k, 64k, and 32k files taper off ‘gracefully’ at their respective top-ends, whereas the cutoff of the VBR files is much more abrupt. I suspect that reflects the encoding algorithm in iTunes.

When I run the test at 256k I top out between 17 and 18 seconds. (I play through decent headphones, close my eyes, and hit ‘Pause’ when the sound disappears.) The same holds for 128k. At 64k the sound disappears (for me) around 15 seconds, and at 32k around 8 seconds.

I also re-encoded the 64k and 32k files with Variable Bit Rate enabled, and the high end warbles just like it’s coming out of a steaming water kettle whistle. Interestingly, even though the top end of those samples lack fidelity to the true wave form, I can still hear sound up until 17 seconds again.

Similarly, with Naxos’s large selection of (mostly excellent) 20th century American works, there’s a need for me to separate early-mid 20th century tonal/Romantic American (Ives/Piston/Copland) from the later works that are more atonal or post-serialism. My musical mood isn’t always for a particular composer so much as for a particular era and style…but then as noted above, such a division leaves someone like William Schuman in an odd spot as he straddles both sides of such a dividing line. (side note, I love the Schwartz cycle you’ve released over the last few years.)

Usually in this case I just pick one side or the other and adapt.

In addition (not mentioned here) I use the “grouping” field to associate pieces together – e.g., each movement of Beethoven’s 5th is “Symphony 5″. If there’s an additional subtitle, it goes here in single quotes, “Sympony 3 ‘Eroica’”. This way my software can assemble m3u playlists grouping things together for a single-click to trigger playing the whole set in order (the playlist title ends up “Composer (last name) – Grouping (the work) – Conductor (last name).m3u”.

Your first question highlights a typo in that footnote, now corrected. What I meant to say is there’s no good reason to NOT use VBR encoding. This is why it is pretty much ubiquitous. Sorry for the confusion.

As to your second question, it’s complicated. There’s more to compression than missing frequencies. A CD and a high-quality MP3 will both give you 16-bit, 44.1kHz playback, but they’ll store the music in different ways. The CD takes a brute-force approach to storage that guarantees the almost exact reproduction of the input. This is the way CD players like to get their data, but on a computer it is inefficient. Silence takes up as much space as complicated music, and that can’t be necessary. The MP3 will try to give you 100% of the audible data in a fraction of the space.

Rather than looking at the difference between different MP3 bit rates, I’d suggest comparing each bit rate to CD. The thing to do here is to try to tell a CD (or FLAC file) from an MP3 when you don’t know which is which. No peeking. If you can’t tell the difference, then there’s not a lot of point in you storing those big files.

It’s tempting to think we need to get the best-possible quality now, because we plan to become more discerning listeners over time. That’s a perfectly good approach so long as your hard drive won’t fill up before you next decide to buy a computer.

In truth, though, our hearing generally gets worse, not better. If you can’t hear the difference on a moderately-priced stereo now, you will probably never be able to hear the difference, even if you buy a much more expensive stereo later.

I hope this helps,

-Andy

I have two questions, You said “I haven’t addressed variable bit rate encoding in the main body of this post because it’s pretty ubiquitous now, and there’s no good reason to now use it.” Can you elaborate? If there is no good reason to use it why is it ubiquitous?

How does this article and the one on Frequencies relate to a statement like “full CD-quality (16-bit, 44.1kHz)” as you mentioned to your reply to Julio? If I decided that all the frequencies I can hear will be captured in a 192 kbps MP3 and I can’t hear a quality jump between 256kbps and 320kbps then how does that fact relate to (16-bit, 44.1kHz)?

You can get a full CD-quality (16-bit, 44.1kHz) download of that album here:

http://www.theclassicalshop.net/Details.aspx?CatalogueNumber=NS%200474

For independent labels, at least, there are several classical specialist download stores offering lossless and hi-res downloads, and we try to send them the highest-quality masters we have. We handle digital distribution for a number of important audiophile labels, and I can assure you, they’re not dragging their feet. There are, though, some commercial hurdles to overcome.

Unless something is scheduled for release on SACD or some other high-quality format, we’ll normally record it in 24-bit at a high sample rate, and convert it to 16-bit, 44.1kHz when the album is mastered for CD. To release high resolution digital tracks, we’d have to go back and remaster them from the high resolution source, which would be a big one-off fixed cost.

When there’s a big enough demand to justify these costs, that’s exactly what we’ll do, but for now the demand is fairly small, and we’d have to charge too much for them.

In the meantime, we’re looking at ways to cost-effectively bring you the highest-possible quality from our new releases.

-Andy

Absolutely true in theory, but in practice this hasn’t happened yet.

I actually ended up viewing this blog, and this post, completely by chance. What set off the string of events is somewhat relevant.

I had just finished listening to the Toscanini 1953 recording of Dvorak’s 9th symphony on LP. I own it as part of a collection released decades ago by The Franklin Mint called “The 100 Greatest Recordings of all Time”. Looking for a 24 bit / 96 or 192khz download of this piece, I eventually found something from Naxos, but it was only available as a standard mp3. The LP still sounds better, despite its status as a senior citizen.

The computer audiophile community is growing, and it’s not necessarily limited to people with huge budgets. Good DACs are getting cheaper, as are decent speakers. We are starved for good downloads and it seems that all the content owners are dragging their feet on this.

It will take a generation before all the individual constraints (storage, bandwidth, DACs, shielding, speakers, etc) are insignificant enough that everything can be listened to at the limits of the best auditory systems. For now, us fussy types will have to endure the subtle (and sometimes not so subtle) taunting from people like Mr. Kattleman.

I will say, though, that although the PCM data can only describe one data point for each of the extremes of the wave and so records no information about the shape of that wave, the DAC won’t reproduce it as a simple square wave, because it does a whole bunch of filtering on the output signal.

I don’t understand your second point.

It sounds like your DAC was a worthwhile acquisition.

The sensible thing to do for any audio test is to use whatever you’re going to use to listen to the music in a real situation. If you do all your listening on your laptop, you’re not going to miss frequencies it can’t reproduce. If you used your laptop to pick a compression codec and then switched to your DAC, you could be quite disappointed.

Conversely, if you went to a super-quiet acoustically-controlled environment like a mastering studio to pick a codec (or a piece of equipment), you might be taking home rather more firepower than was useful. People would do well to remember this when they’re sitting in the demonstration room in a hi-fi shop.

There have always been compromises. The big difference with digital audio is that it gives you, the user, the opportunity to decide for yourself where you want to draw the line. This is an important point, and I should’ve put it higher up.

*They’d also wear out quickly, but then if LPs had been designed to turn at 78rpm, they’d probably be made of something else.

When taking the sound through an external DAC, the wave is fine. Well, presumably it is… at least the progression is smooth and after 27seconds I just assume it continues.

As a kid I could hear 20k, when my uncle brought out a test-vinyl. But with with age, the top frequencies get lost and I’m now down to 18, more likely 17k.

I don’t know what percentage of MP3 consumers owns a traditional stereo system. I don’t have one (although there are lots of speakers in my house). There are, though, lots of people who do own a stereo and don’t download music. If that’s because they love CDs and they’re happy to get their music that way, then of course that’s fine with me. If it’s because they don’t know how to get the music out of their computer, then that’s something I’m well-placed to help them with.

As for the last question, you’re quite right that there’s no point in owning something you can’t hear. It takes quite an investment of time to encode your whole CD collection, though, and you don’t want to do that twice. It would be a real shame for somebody to hook up their computer to their stereo only to discover they could hear all sorts of encoding artefacts. Better to discover them now, and encode everything at a rate that sounds good to them from the outset.

It’s nice to hear from you.

You’re right, of course, that it’s easier to hear these things when they’re louder, and that the ear’s frequency response is not the same at low volumes as at high ones.

When we digitise sound, though, we can only record up to a limited amplitude (described as 0dB full scale, or 0bDFS), and that maximum amplitude is the same for every frequency. If you turn it up until the low notes are as loud as you can bear, and the high stuff still isn’t audible, then it is going to be very difficult to construct a musical recording in which these frequencies are audible.

As for the lower frequencies you attribute to aliasing, I think they’re probably created during playback, somewhere in your hardware. I suggest this because:

1) There’s considerable variation in MP3 encoders (I used the Fraunhofer encoder, the LAME encoder is also popular), but the MP3 specification doesn’t leave much room for innovation in the way the encoding is turned back into sound, so everybody should get the same PCM waveform out of the file, regardless of the software they use to decode it. What varies is the way that PCM stream is turned into sound, and that happens in your sound card.

2) I was concerned that such problems might arise, so I looked for them: I’m effectively using an MP3 to show you the things some MP3s can’t contain, so I needed to be sure this MP3 did contain them. To check, I converted the MP3 back into PCM audio, reversed the polarity*, mixed it with the original source file, and checked that they cancelled out. I also visually inspected the waveform to ensure that it was the same shape and without the sort of obvious harmonic distortion that would create additional audible frequencies.

3) We’d expect everybody to hear the same thing – at least, we’d expect the people who heard all the way to the top to hear the lower frequencies.

4) Aliasing occurs at half the frequency of the sample rate. The sample rate here is 44.1kHz (the same as for a CD), so we should be good up to 20kHz. You certainly shouldn’t experience any aliasing if your hearing cuts out before 15kHz. Since aliasing is a product of the sample rate, it would happen with lossless files too. Here’s the WAV file of the 20-20k sweep.

http://blog.naxos.com/wp-content/uploads/audiocheck.net_sweep_20Hz_20000Hz_-3dBFS_30s.wav

You might also try burning this to CD and playing it back on something else. That would tell you if what you’re hearing is a product of the file or your DAC.

Finally, let’s talk about compression in radio. I’m guessing you know this, but there’s two kinds: dynamic compression (where you make the loud bits quieter, normally so you can make everything else louder) and then there’s lossy compression, where you make the files smaller to save space.

These two different kinds of compression work in totally different ways to achieve totally different effects.

Radio stations often apply dynamic compression to everything they broadcast. The idea is to make all the music (and talking) audible, even if you’re in a noisy environment, like a car. If it’s done right, this is helpful to most listeners, at the expense of a slightly claustrophobic feel to the sound for those listening in very quiet places. Done wrong, it can make all the peaks sound crunchy and unnatural, create sudden quiet moments after each loud peak, or make everything sound boring. Radio engineers would do well to remember that most modern popular music has already been compressed and limited to the point where there’s not much mileage left in this technique.

Digital radio stations (and this is particularly noticeable with American satellite radio stations) use very low bit rates to fit many channels (or many listeners) in a small amount of bandwidth. If it sounds bad, it’s because they’ve done too much of it.

Thanks for reading.

-Andy

* Don’t try this at home, and if you do, don’t cross the streams or you might blow the Flux Capacitor.

http://en.wikipedia.org/wiki/Equal-loudness_contours

I listened quite loud and I can detect what seem like “aliases” in the sound. That is other lower frequencies that have been created by the digitisation of the signal. Of course what is being sent over the internet in your test is a digital sound signal.

http://en.wikipedia.org/wiki/Aliasing

I cut out at about 11-12 kHz but have to have the volume high for this. Many years ago I stopped hearing tape hiss and listened to cassettes with the Dolby off. Still do. ( “What’s a cassette?” Young person) Moi, born in 1947.

What is really annoying is the way that digital rock music stations which use lower sampling rates and high compression make the music sound raw and distorted at the top end.

When it comes down to it, if you enjoy the music, that is what is important. When I was younger I used to say never mind the music listen to the reproduction. Now I just enjoy the music. I no longer have a golden ear.

http://en.wikipedia.org/wiki/Golden_ear

The way these codecs work is to describe the waveform in terms of lots of little bits of sine waves of different amplitudes and frequencies. It’s only ever an approximation, but given enough data, it can be a very good one.

Where it always struggles is with sounds that aren’t obviously “musical” – applause and very percussive sounds present particular problems, because they consist of sharp impulses rather than continuous waves.

The idea of describing the shape of a line with an equation instead of a long list of X/Y co-ordinates isn’t inherently inferior (indeed it has some powerful advantages) but its weaknesses are complex and less predictable.

Because of this, because of the many variables involved, and because the constant developments in encoding algorithms, it seems the sensible approach is to judge its performance on a fair assessment of the quality of the output, rather than on an a priori supposition of what ought to work.

Frequency response alone isn’t a good way to pick a bit rate or file format. I offer the “how high can you hear test” as reassurance, not in an attempt to convince anybody to change their behaviour. Such a test can cast serious doubt on the idea that the absence of a particular frequency is deleterious to your experience of recorded music, but it certainly can’t prove the adequacy of a file format.

That’s great. Do you know how far above 20kHz? Here’s a test that starts at 22kHz. http://www.audiocheck.net/audiotests_frequencycheckhigh.php

Perceived sound quality is less about how high the harmonics go, and more about what happens up there in terms of tonal balance and harmonic distortion. The later is a widely misunderstood term: it’s not like ‘fuzztone’ or other kinds of distortion we might associate with rock guitar. ‘Harmonic Distortion” means the reproduction creates overtones at frequencies or levels that did not exist in the source. These overtones are way up at the edge of our hearing so we don’t perceive them as sounds-in-themselves, but as subtle differences in the character of the sound. The musical term ‘timbre’ is largely (but not exclusively) about differing overtone structures. So harmonic distortion affects ‘timbre’. And it’s not strictly a numbers game where less is better. Tube amps, for example, produce a LOT more harmonic distortion than solid state amps, but they tend to add overtones the ear finds acceptable if not pleasing, while solid state devices tend to add overtones at frequencies we experience as harsh. And the distortion produced by electronics is small compared to the distortion produced by transducers (e.g. speakers or headphones) which is more often than not also of the harsh variety.

Now, I have no idea how any of this is affected by codec algorithms, but any high efficiency lossy codec like MP3 isn’t just making a digital sample of what was there in the analog signal. It’s making choices about what to keep and what to toss out, and I would guess this is not just a matter of topping off the high end, but could also possibly alter those harmonic structures????

Using your test I can hear the entire thing quite well.

As I recall, the statement was that 80% of people hear 20-20 only. Not 100%.

It’s certainly a valid point; one worth mentioning also for speakers that claim to go to frequencies above 20 kHz.

I’m not sure if you’re saying that the availability of legal downloads is somehow correlated to online piracy. Obviously both require the Internet, but there’s some evidence that music unavailable for legal download is more likely to be pirated.

As you say, there are stores offering lossless recordings. You can see more than 1700 Naxos recordings available in FLAC here at The Classical Shop. This is an alternative to buying the CD but it often seems as if people would rather have the disc.

I gentely disagree that FLAC is not widely supported format.

Pretty much every single android phones can play FLAC (thanks to Samsung) and with 3rd party (which is usually free) iPhones also can play FLAC files no problem.

For dedicated music players, even cheap $40 Sansa Clip can play FLAC files these days.

It’s been a long time (read : more than 5 years) since both Itunes and WMP can play FLAC with plug-in, and other major music players such as Winamp, foobar2000 support FLAC playback for a long time.

Heck, even some of bleeding-edge cloud services support FLAC files. I was very surprised when Google Music Beta automatically detected my 24bit FLAC files and stream them into my android smartphone. Probably it is downsampled, but it is very welcoming support from a leading tech company.

Hardware-wise, I believe FLAC has been fully supported since as early as 2005…. what -still- lack are accessibility and availability of the contents.

At last, Linn Record revealed that once they started high-def music download service, (SACD/DVD-A/Bluray quality) the overall percentage of people choosing download was more than 60% at the first day of the launch. Currently the figure is more than 90%. Remind you Linn also offered mp3, normal-quality FLAC, physical CDs and even Vinyl, but 9 out of 10 chose high-def FLAC.

People buy CD, only because they have zero alternative to get high-quality music otherwise.

However, I can’t get it to play Naxos.com for which I have subscription (before I got the NP30 I had been listening via the NaxosPlayer in Firefox7.0.1 on my laptop computer).

Does anyone have any hints to offer on how to make this work – it may just need to be able to handle the Naxos subscriber log-in process. Or do I have to accept that Naxos.com won’t work with my media player?

Thanks,
Jonathan

You raise some good points here. You can currently stream music from:

- Naxos Music Library
- ClassicsOnline
- Rhapsody
- Napster
- Spotify
- Classical Archives

You can currently download music from:

- iTunes
- Amazon
- ClassicsOnline
- EMusic
- Rhapsody
- Napster
- Classical Archives

In time, somebody will have to resolve the high-quality downloads issue. It’s tricky for a mainstream retailer like iTunes, though, because the files are sometimes 10-20 times the size, and if you’ve got 20 million tracks to store at that quality, all ready to go, on servers all around the world, those costs start to mount up. Right now, I think a lot of the people that might buy higher quality downloads are fairly happy with CDs.

I do think there’s an opportunity for us to improve the FLAC offering on ClassicsOnline. It’s a shame that FLAC isn’t more widely supported by the popular portable players. We looked into offering a more universally supported format like WAV. I think we could deal with the fact that the files are 40% larger, but the real problem was that they don’t support ID3 tags, so you wouldn’t get any metadata with your music.

-Andy

The numbers here are all gross wholesale revenue, so I haven’t included our the retail margin from our own stores. This seemed like the best single measure of the amount of business each store does.

The difficulty with any other metric is that no two stores work exactly the same, even within the broad categories of streaming and downloads. One store might define a “unit” as a track or album, another as one second of music. Some generate (and share) revenue from non-musical sources like advertising or unredeemed store credit, and cloud service revenue complicates things further. Some streaming services pay a per-stream minimum, others per-second or per-user. Some just add up all the cash and share it out.

When it comes to selling digital music, there’s some ordinary apples, a few pears, a whole lot of grapes and even some orange juice, but at the end of the day, it’s all fruit. If we were selling it all, we could measure it by weight, but since some of it is only rented (and I can’t think of a fruity parallel to this bit), we have to measure it by what we got paid.

You raise an interesting question though – how many times does the average download get heard, and would people be better off subscribing?

For me, it doesn’t make much sense to even try to work it out. We might find that, on average, one group of people (subscribers or downloaders) ends up paying more, but nobody is average, so that wouldn’t tell us what was right for any one person individually. If you like to listen to a small repertoire over and over, you’ll pay less by downloading it once. If you like to listen to something new every day, you’ll save money with a subscription, but it won’t be quite as portable.

Like a lot of people, I do both.

From a business standpoint (and I think this is what you’re really asking), I can see how it makes sense to compare how much we make from the sale of a CD, a download and a subscription, and to compare how each of these gets used. The trouble here is that most people don’t need music, so demand is flexible. We can’t assume that all subscriptions replace sales, or that every download represents another CD gathering dust in the warehouse*. There are many factors at play, all interacting with each other. Every time we try to construct a mathematical model of what’s going on, the assumptions start to pile up until we don’t really have any faith in the results. If we really want to know what will work, we have to experiment, on a large scale, in the real world.

-Andy

* This is a figure of speech. Our warehouses are actually very clean. Our biggest warehouse is in Munich, where we have a huge vacuum cleaner like a mini tractor. It looks like a lot of fun. If I behave myself, perhaps they’ll let me drive it one day.

Thanks – very interesting statistics. I’m wondering: are these numbers revenue or unit volume? Because I think you’re sort of comparing apples (no pun intended) to oranges if you graph download and streaming services together. As you know, royalties from downloads differ greatly from those from streaming. It would be interesting to see the download and streaming sites broken out and compared against each other, and to find some data on average plays per download in order to more properly compare the two.

It is excellent, it allows the connection of audio in and audio out using RCA cables, and headphones.

Personally, I have thousands of CDs, many of which are box sets that I only keep in case I might, one day, want to listen to something off them. I have a rule that I’ll rip a CD the third time I listen to it. It saves a lot of time.

One fair day I’ll afford upgrading to a 5.1 system and probably a stand alone blue-ray drive.