Audio here, there, but not quite everywhere

I’m a big fan of online streaming audio. I loved Pandora before it became US-only. I quite like Last.FM. Musicovery has a lovely approach in its interface and mood-based approach. Now – at last – we have Spotify, and another great utility I came across this week, Simplify Media.

I’ve covered Spotify before. I saw a pre-beta version and was very impressed with the immediacy of Spotify’s streaming. Then it went to invite-only beta so I was delighted to receive an invite last week and I’ve been playing around with it since. It has good technical points but I think it’s missing a big marketing trick.

Just to round up the current offerings, with my take on each:

  • Pandora uses the Music Genome Project in which musicologists analyse a song using many parameters, starting from the basic – tempo, style and so on – and then really dig down to whether it’s guitar-based, solo, male/female vocals etc. It’s not very community-based – that is, it doesn’t become more sophisticated through referrals and relies on the possibly subjective analysis of a small community – but I loved the results. Type in Nick Drake and you could spend a wonderful afternoon with the likes of Bob Dylan, Jeff Buckley et al. It just… worked. Unfortunately its business model didn’t and so it had to batten down the hatches in the face of prohibitive licensing costs and is no longer available outside the US. You can try IP-rerouting to trick it, but with limited success.
  • Last.FM is the leader and adopts a pureplay referral system. In the same way you see references on, say, Amazon telling you what other people have bought that is similar to the product you’re looking at, Last.FM infers that, because you like song A, you’ll like song B, because other people with similar tastes liked them too. It’s a great theory and it works well with, say, electrical goods or DVDs, but I’m not so sure the algorithm works as well for music. You can stream by artist or by tag and of the two, tag works best. If I listen by artist I’m often dismayed that Last.FM will give me American soft rock when I want to listen to music that is ‘similar’ to Beck, for example, simply because other people’s tastes straddle the two. But you can specify a tag – eg jazz – and have a reasonably consistent listening experience.
  • Musicovery’s great insight is that you tend to be in a certain mood when listening to music, or want to specify music to reflect your mood or change it. So you can specify whether to listen to light-hearted music, or something a bit darker, and the tempo, choose across many genres, even specify the decade of the music, and be up and running with a very pretty Flash-based interface. It can throw up interesting results – I did not know there was a jazz version of OK Computer, for example – and I like it for that. It truly is music discovery.

So Spotify needs to find its niche within these established players. On first glance it looks very much like a greyscale version of iTunes but is initially blank, which I found quite offputting at first. I just wanted to see at least some initial offerings to choose from.

But type in your artist and it immediately springs to life. And how. It’s incredibly responsive. Click a track and it almost instantaneously starts to stream. It’s very easy to create or share playlists. And you can choose to listen to ‘stations’ made up of genres and, like Musicovery, timelines. 1960s Heavy Metal digs up Steppenwolf’s Born to be Wild, while 1950s Reggae gives… interesting results.

What we have here is essentially an iTunes interface to an online streaming music database. It’s designed from the ground up to live online. And this is where I don’t get part of Spotify’s proposition. As far as I can tell it’s a completely self-contained system. I doesn’t seem to offer any capability to integrate with other systems. For example, you can’t scrobble audio with Last.FM. And, more importantly, there are no widgets, for example for Facebook.

Isn’t this missing a trick? If you’re web-based then isn’t this something you would immediately implement? I can only imagine this is part of the plan, to be rolled out post-beta (assuming it’s not a permanent beta like so many web apps nowadays), or maybe there are licensing restrictions.

I also think another challenge for Spotify – as with all the other systems mentioned so far – is the sound quality. It’s ok on my low-end DigiTheatre surround-sound system in the lounge because that’s not a hi-fi. But when I listen to it on my studio monitors, the limitations of the sample rate become immediately apparent. It’s compressed, there are artefacts, and a like-for-like comparison with my own music library shows how locally-stored MP3s are far superior. Hmmmm.

POST EDIT: Following comments from Daniel at Spotify, I looked at (and listened to)  this more closely and the differences are not as clear as I first thought. Please see my reply on this subject.

This is where Simplify Media could offer an alternative. It enables music sharing across IP, and integrates with iTunes and Winamp. Download the client – at a hefty 14MB it’s not exactly light – set up an account, and share your music. Get up to 30 friends to do the same and suddenly you have access to many thousands of tracks.

Last night, for example, I listened to the Bob Dylan tracks on my friend Paul Borge‘s MacBook Pro, in my lounge. Quite apart from the mind-blowing technical feat here – from his machine, across his wireless network, across the web, through my wireless network and onto my machine – the sound came through with, as far as I could tell, original quality. It sparkled (inasmuch as one could describe Bob Dylan’s whine as ‘sparkling’).

Again Simplify Media doesn’t integrate with Last.FM scrobbling or Apple’s new Genius playlist feature, but you can kind of forgive this. It’s strictly a music sharing system, as opposed to fileshare, and I like it a lot.

What I’d really like to see is a system that enables the ‘DNA’ of the music to be tagged like Pandora, in a sort of ultra-high resolution folksonomy, not just broad categories such as genre; the community dynamics of Last.FM; the clever interface of Musicovery; the agility of Spotify; the quality of Simplify Media; and full integration with widgets across the most popular social networks.

Maybe each of these systems is addressing a different way of listening to music. But, in the same way as we’re getting aggregators of aggregators in Friendfeed, or blog editors that interface across many platforms such as Live Writer, I’m wondering whether someone, somewhere, is going to come up with a system that has all the advantages of each of these approaches, and none of the drawbacks.

Me? I’d have been quite happy with Pandora to be honest. Maybe I should move stateside.

35 thoughts on “Audio here, there, but not quite everywhere

  1. I quite miss Pandora, although I’ve found that the larger my library on, the more I like it. I just can’t quite bring myself to pay to organise separate playlists of music that’s locked into existing for all eternity – probably more of a mind shift required!

    Have you tried the ‘twitter for music’? I quite enjoy it as a bit of a change – it’s essentially like a crowdsourced John Peel show collecting music from your friends and contacts…quite be really good for discovering new music!

  2. Brendan,

    Would love to get feedback on the sound quality issues you’ve been having. Spotify is streaming the music at 160kbps Ogg Vorbis q5, which should be equivalent to a lot of the content on iTunes. We are doing volume normalization, which right now causes Spotify to have lower volume than say iTunes. However, this should not create the artifacts you are experiencing. Can you provide us with a couple of songs that you’ve been experiencing this with?



  3. OK, here are some comparisons.

    I’ve recorded a passage from Flaming Lips Yoshimi Battles The Pink Robots – a highly compressed track anyway, as is the whole album.

    I’ve channelled these internally into my soundcard – a Hoontech DSP24 Value card, sweet AD converters – into Wavelab Lite as 16-bit 44.1 kHz wave files. So it’s purely within the digital domain, and saved as uncompressed wav files of around 5MB each.

    1 Yoshimi iTunes m4a.wav – is recorded from an iTunes m4a file which was originally encoded at 128kbps. When I said ‘iTunes’ I was referring to locally stored files converted from the original CD.
    2 Yoshimi Spotify.wav – is recorded from Spotify
    3 Yoshimi Last FM.wav – is recorded from Last FM

    They do have different playback levels, so just for fair comparison, I also normalised the recordings to -1db:
    1 Yoshimi iTunes m4a NORMALISED.wav
    2 Yoshimi Spotify NORMALISED.wav
    3 Yoshimi Last FM NORMALISED.wav

    I wasn’t able to simulate this through Simplify Media because my friend wasn’t online.

    So, I’m going to put my hands up here – and edit my original blog entry. The difference is pretty small. To my ear the iTunes recording is slightly clearer. There is a slight sibilance to the Last FM and Spotify versions, which you can hear in the ‘s’ of, for example, the word ‘robots’. I’m also minded to think that the bass drum isn’t as defined in the Spotify and Last FM versions as it is in the iTunes version.

    When I said ‘artefacts’ I was probably being too harsh because this sibilance and low-end woolliness is what I was describing. It was a catch-all term to describe ‘things you can hear which shouldn’t be there.’ Let’s be clear – I wasn’t talking about audio glitches or pops.

    But actually this is indeed slight. I think that if you listen to the non-normalised versions you get a flavour of what I was originally hearing – the listening volumes are different and I admit freely that differences in apparent listening volume are a basic error to make. Normalisation removes that difference – as would have simply turning up the volume on my system! (and in fact this is why a lot of contemporary music is highly compressed because louder is often perceived as better).

    So, Spotify, you win on all accounts! I’d like to point out that:
    1. I was being over harsh in what I said.
    2. You were savvy enough to pick up on my blog post.
    3. You were even more savvy to pick me up on it, publicly on my blog.
    4. You’ve been vindicated.

    This is a perfect example of good, transparent, proactive, public engagement. You rock. And at least I’ve made good on what I said too. As I say in my About page: If I make a boo-boo, I will own up to it.

    PS I would still love to see a widget or two… 😉

  4. Brendan,

    Glad to hear that there wasn’t any artefacts. We are definitely working on how we can do volume normalization better.

    In regards of widgets, it’s an area where we hope to make lots of progress. We’ll make attention data APIs available for third-party developers who could easily hook us up with or build a bunch of other cool stuff. Meanwhile you can look at, and to see what cool stuff people are already building.



  5. Pingback: » Catching up…

  6. Daniel:

    I immediately acquired the monthly subscription of Spotify without testing it first, and the problem I’m facing now it that the result of level normalization of the content is preventing me from enjoying the music on my laptop since the internal headphone amp is not powerful enough to power any of my headphones (AKG 324P, K240S & K601) while other players (with no normalization) provide just enough output to get a decent level. The obvious solution is to let the consumer choose if he/she would like to use the feature or not. I assume that the feed is in original level and that the player extracts the RMS information embedded, and then adjusts the output level accordingly. I’ve read other posts on forums where people are complaining about the “quality” of Spotify but the real problem is more likely to be the level difference. This could become a serious problem if you do not inform the users and give them options. If this problem I not addressed, I’m sad to say that I need to abandon this service since I’m working mostly with my laptop. I really think Spotify could be huge but don’t let it fail because of this!

    Sincerely Jens

  7. Are you sure other players don’t use normalisation? Which ones are you referring to specifically?

    I can understand *why* players use this, in an attempt to remove the differences in subjective volume levels between tracks across different artists, albums, genres etc. I believe iTunes’s ‘Sound Check’ feature does something similar.

    Having said which, I’m not convinced it works that well in practice. I remember doing this manually for a load of tracks some time ago, and quickly realised that volume level is very subjective. If a track is compressed anyway, then it will sound ‘louder’ at the same level as an uncompressed track. Try listening to a classical piece at the same level as almost any contemporary mainstream piece (or, indeed, anything contemporary with anything, say, a decade ago – producers compress the heck out of everything nowadays).

    Also, while normalisation *shouldn’t* alter the character of a track – it just lifts the waveform rather than altering it – in actuality, it does. Or rather did, when I used it. So perhaps something is awry with Spotify’s normalisation algorithm if what you say is true.

    I wonder whether Spotify really is ‘normalising’ at all. Surely compression would cure all these ills – both leveling out the balance, and ensuring a decently ‘hot’ signal? I only ever use normalisation when I’ve got a slightly naff recording and need to make the most of the headroom available. If I want to smooth out differences in volume, for example in a vocal piece, I’ll use compression.

    I dunno. I don’t work for Spotify. Perhaps I should. And you should definitely let them know about the issues you’re having.

  8. If normalization is applied correctly, the loud tracks will be lowered in level since you can’t make the quiet (dynamic) ones louder without applying limiting or compression (or both) and this option is of course out of the question since you then alter the sound of the recording. The problem that appears the is the fact that some songs, like early jazz etc. has a very low RMS level and if a modern rock song with a high RMS level (closer to -10dB rather then -22dB) is played, the player then lowers the volume of the rock song by 12dB if the process uses the jazz tune RSM level as reference. Further more, this results in the fact that the consumer with a normal soundcard (16bit) will hear the music in 14 bits instead of 16 bits resolution because the level is lowered 12dB in digital domain before the DA-converter. To compensate for the lower playback level, the listener will raise the volume on the speaker system and thereby raising the noise floor of the whole system as well. Some (like me) can’t raise the level any further and will return to WinAmp, Media Player or others, if there is no solution to this issue.

  9. I just might and yes, I’m an audio engineer with a lot of experience with automated processing of audio in regards of level…


  10. I can tell! It reminds me of the conversations I used to have on the Computer Music forum (now on Music Radar but I haven’t visited for ages). It quickly becomes apparent whether someone is a ‘noob’ (ie “What’s better, Mac or PC?”), or someone who just likes playing around with this stuff in their spare time, or really does know what they’re talking about at the engineering level. 🙂

  11. I have sent a mail to Spotify with a link to this blog and I hope that they realize that the best solution is to let the client decide (as always) whether they want to use normalization or not.

  12. Did he make any comments?

    Another problem with this is that if you do have enough juice in the headphone amp to power your cans, your in for a horrible surprise if your listening to Spotify while surfing the web and suddenly there’s a page containing audio that starts to play at a much higher level than Spotify and I usually want to have the level up fairly high.

    I really hope the next version will let the user decide if they want to use normalization or not.

    Sincerely Jens

  13. I finally got a reply from Spotify but they didn’t say much: “…this is an active topic of conversation internally but other than that I can’t say much more.”

    I wonder if the level normalization is a cute excuse for them to be able to play the commercials louder than the music in order to be able to pleas their clients…? Otherwise the solution is simple: let the user decide (or at lest the paying ones).

  14. Is the music playing louder than the commercials? In fact, are there commercials? I can’t remember – when I last listened there weren’t any, but I’ve been offline so long this could have changed.

    If the commercials are louder then how would this differ from commercial radio? I think the big difference is that people feel more ownership with online – that is, they feel they could/should be able to remove adverts in a similar way to blanking out banner ads – and in fact, if there are ads, there are also plenty of ad-free alternatives to choose from.

    My take on this is that the normalisation is simply to balance out volume differences, like the iTunes feature. What puzzles me is why Spotify therefore sounded *quieter* than the alternatives?

    This could go on and on. But well done on getting a response Jens. If nothing else this should tell Spotify that there is discussion outside of their immediate online sphere and they should at least be listening to it, if not interacting with it.

  15. In the free version there are commercial announcements between (or maybe in) songs. Since I’m a paying costumer I don’t have much experience of it but I have seen posts where people are complaining about the level of the commercial is to loud so…
    Spotify shouldn’t be lower in level than iTunes if “level normalization” is used in both apps (or at least not much depending on the reference RMS-level).

    I really don’t think that any of my fears are real but one gets curios when there seems to be some issue surrounding this matter. A user option to turn the feature off is technically very easy to implement unless the have lowered the level of the songs destructively (which I assume they haven’t). The most likely reason for not adding the option is the fact that commercials will be lower in level than the music unless it is “correctly” produced.

    Future will tell and I hope that my theories are way off and that the next version of Spotify will let the user decide if the want to use “level normalization” or not.

  16. How interesting – thanks for passing on. And yes, this is a prime example of how Spotify can and should listen to their audience, and should take action. Let’s see how this winds out!

  17. Pingback: FreshNetworks Blog » Blog Archive » Spotify and the Clones of Dr.Funkenstein

  18. Pingback: Catching up… | iphones |

  19. Pingback: Catching up… | applications |

  20. Interesting question!

    I actually wrote all this before it became trendy. I loved Spotify when I first saw it quite a while ago, and got onto the beta programme.

    Since then it’s suddenly exploded onto the scene, so this blog post has atttracted quite a lot of attention. That’s an interesting point about posts: issues come and go but posts remain online, so they occasionally pop back up in the stats.

    To answer your question however, usually I write what I like (prerogative of being a blogger I guess), and that is usually just something I’ve been thinking about recently, or a response to something I read or heard.

    Occasionally however there are ‘scheduled’ posts. I try (really, I try) to update the PR Friendly Index monthly, but it’s becoming very difficult to maintain and I’m not sure what real value it serves me.

    Occasionally too, I post about something because somehow, I know it’s going to grab attention. My response to the BBC viral video ended up being quoted on the BBC website (but not, alas, linked from). I knew that, if I could get my blog post up there quickly that morning, it would pop up in discussions. The fact it landed on the BBC website was an unexpected success however!

    Really, I should have a schedule, and that’s what I’d advocate. Think about what you want to post about but have a schedule for upcoming events, seasonal posts, reactive posts, etc. In other words, do it properly, not in my ad-hoc manner.

Look! It's a comment field!

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s