Feature Friday: Track Stations

Our portfolio company SoundCloud quietly launched something this past week that has quickly become my go to music discovery experience. It is called “track stations” and the concept is not new but they have applied it a bit differently. Like Pandora, you can enter a song you like and get a radio like listening experience.

The difference is that SoundCloud has something like 110 million tracks in it versus something like 2 million tracks in Pandora. That’s important because it means that there are way more tracks to start stations with but even more importantly, the track stations give you access to a long tail of content that you can’t really access any other way. Using algorithms and data science to take you from the head of the curve content to the long tail is an interesting idea across many content categories, but it is particularly interesting in music and podcasting.

I started a track station one morning this past week in my car and within 30 minutes, I had added five or six tracks to my favorites. And I didn’t know any of the artists who made any of the tracks I favorited.

It works like this. You open the SoundCloud app on iOS or Android and search for a track you know and like:

2016-02-05 07.46.30

Then you click on the track to start playing it:

2016-02-05 07.47.07

On the lower left are three dots. If you click them, you get:

2016-02-05 07.47.12

And if you click on “start a track station”, you get:

2016-02-05 07.47.18

If you go to your collections in the app, you will see all the track stations you listen to regularly:

2016-02-05 07.49.54

But here’s where it gets really interesting. You can use the same technique to find new podcasts to listen to. If you start a track station with a podcast you like, like this one:

2016-02-05 08.21.14

You can swipe through to find similar podcasts like these:

2016-02-05 08.00.25 2016-02-05 08.00.32 2016-02-05 08.00.38

So that’s it. If you have SoundCloud on your iPhone or Android, find a track you like and start a track station. It’s a great way to find new things to listen to.

Right now, track stations is only available on SoundCloud’s mobile apps. It will be coming to their web app and other apps soon.


Comments (Archived):

  1. Sebastien Latapie

    This is huge! Music discovery is such a huge opportunity. iTunes does such a terrible job with it. I’ll definitely be using this feature.That being said, while SoundCloud is improving on discovery, they should make it easier for me to re-listen to what they’ve helped me uncover! I wish I had the ability to search through all the tracks I liked. There is nothing more frustrating than scrolling through hundreds and hundreds of likes just to find that one track from a few months ago.

  2. Pranay Srinivasan

    To start a radio track with podcasts is genius! I dislike picking thru podcasts selectively..Also would USV / Fred ever start a biweekly podcast? Would love to hear a good natured unstructured conversation about “This week in AVC” – Also could post well to Audio Tuesday or something

    1. fredwilson

      i did a podcast back in 2005 and 2006. i dropped it after a few years

      1. Pranay Srinivasan

        Maybe your own channel now would be awesome for us to hear – AVC radio!!

        1. Michael Elling

          Fred doing the interviewing of notables instead of the other way around. His questions. His terms.

          1. Pranay Srinivasan

            Yessss and his agenda!!Sent from Outlook Mobile

          2. Lawrence Brass

            Great idea Pranay, Michael… But I like more Fred on video. Not many people have the ability to perform well on video and he has it. A monthly Fred hosted video interview would be really cool.

      2. Lawrence Brass

        Video killed the radio star..

  3. Nathan Guo

    Looks interesting. I listen to a ton of podcasts, but find myself having to curate very manually as individual episodes frequently vary substantially in quality. I feel like the duration of a podcast vs. song (30 mins vs. 3 mins) makes podcast curation 10x more important, but also substantially more difficult.

    1. sigmaalgebra

      Yes, curation is important work and not always easy. So, in the world of search, discovery, recommendation, also important is curation. Then when have a good curation, might want to send that to others.It’s in the plans and the work queue!Sure, and there’s another one — notification, so that to keep up on what is new and belongs in your curation, day by day you don’t have to just keep returning to poll for the new stuff.Can also have subscription, which can be a little different from notification.So, the list of basic functionality is search, discovery, recommendation, curation, notification, and subscription. Yup.Yup, done automatically in real-time, have a case of filtering. IIRC some Twitter users at AVC have been interested in filtering.For entertainment, should be good. For some business purposes, could be really important.

      1. Lawrence Brass

        Current pseudo-AI driven curation and recommendation is getting better, but there is something wrong about it that have always puzzled me. It doesn’t seem to evaluate or remember past responses or choices.Take for instance Google Ads/Analytics: As google scans my main gmail account and I receive hosting bills there monthly, G-Ads thinks that I am a hosting buyer, so every month, specially after receiving the bills, it floods my browser with hosting ads, even from the same hosting providers I use. The thing is that I *never* click those ads, but G-Ads seems to ignore that and keeps insisting on it month after month. So I assume it analyses a very narrow behavioural window and that it doesn’t care about past behaviour.We humans like other humans to remember things about us, it can be remembering positives or remembering negatives. If I noticed what wine variety you chose the last time you visit, perhaps I will offer that as a first choice this time. I won’t offer you a variety you rejected before and maybe I will invite you to taste a new one I have never offered you before.The way to achieve, as @TwainTwain often mentions, ‘AI that cares’ is to train it carefully on the nuances of human behaviour. Probably AI will never really care about us, but it can cheat us to believe it cares, and that starts with remembering what we like or *not* in very subtle ways. Today its still too coarse.So if your are designing and implementing automatic curation and recommendation systems, save people responses somewhere and lower the probability of a future positive response to suggestions he/she ignored in the past.Sometimes the only reasonable thing to do is to walk away, do like HAL and say ‘this conversation can serve no purpose anymore..’Good bye.

        1. sigmaalgebra

          Thanks for the input.For Current pseudo-AI driven curation and recommendation with “pseudo-AI” you just went to the front of the class!ForProbably AI will never really care about us,the real situation, IMHO, has nothing to do with AI and, instead, is simple and elementary: The software designers, writers just have more work to do, common, ordinary, garden variety, simple-minded, vanilla pure, software design work, nothing more or less. I.e., so far their work is not very good.The design of my Web site includes user IDs, passwords, and logins. The relevant columns are in the database schema that is running. Putting up a Web page to permit logins is totally standard Web site art and, with the available tools, fairly simple, and with current Web browsers can be quite easy to use because the browser is willing to recognize the domain name of the Web site and, for it, save and use the ID and password. A good dialog for handling IDs, etc. is nicely worked out and popular on the Web and well understood by most users.But for now, I have not implemented IDs and passwords, and that is reasonable for now.E.g., at Google and Bing, can do a search and get results without doing a login. And if want to keep the results, then keep, say, the Web page with the results.Yes, with a curation, my site should offer a way to keep that, return to it and refine it, send it to someone else so that they can refine it, etc.Sure, maybe a CFO does a search to get up to date on Sarbox, gets a curation, e.g., a reading list at about the right level for his background and his purposes, to keep up has a subscription, say, redo and send the results once a month, and asks for notifications, say, send when the Web site suddenly happens to find something that looks especially relevant.Then the CFO may want to send the curation to each of his worker bees so that they can revise the curation for themselves, etc. So, sure, this way maybe my site gets happier users, a competitive advantage and barrier to entry, some lock-in, some switching costs, and some virality.That’s all in the design and the work queue! At least most of the needed columns are in the database schema.It is mostly for such future features that I am using relational database, e.g., SQL Server.There is a point about a search and curation for a CFO about Sarbox: The content is based mostly just on text, yet keyword/phrase search doesn’t work well. Broadly the reason is the CFO is interested in discovery and recommendation for what he doesn’t yet know exists and not in specific content he knows does exist. Then, what the CFO wants is based on the meaning of the content, and that is next to impossible to characterize at all accurately with just keywords/phrases. Also relevant is his level of knowledge, and more. Well, for various reasons, it’s possible to get the CFO a curation with the content with the meaning he has in mind, and my work is aimed at that. And, similarly for recorded music, most movies, many video clips, most still images, some Web cams, the artistic taste the user has in mind.Yes, my Web server log file includes some of the usual data on each user, e.g., IP address, but so far I’m not doing anything with that. The intention I have in mind is for (A) some descriptive statistics on the set of users and (B) maybe some issues of site security.So, for now, think of the upside: My site is good on user privacy!In particular, so far my site is making no use of Web site cookies!If my site can tell you what movies, recorded music, still images, pod casts, video clips, Web cams, etc., you will like, then it would seem that the site could also know what ads you will like or not like. Yup, for my work, at least as I intend, all true.So, showing ads, some unique, hopefully effective, ad targeting is in the design and the work queue, and then a good solution to what you said about ads — don’t show you ads over and over you clearly don’t want to see — should happen as a special case. E.g., effective ad targeting doesn’t want to show you ads you don’t want to see!But, for the searching, recommendations, getting a curation, etc., each user effort (session, whatever want to call it) is independent of all the rest. E.g., just because the last five times you used my site to get recommendations or curations for some movies and concentrated on, say, action-adventure or men at work won’t influence your next session where you are looking for some Christmas movies for your nieces and nephews or a romantic tear jerking movie so you can get your arms around your girlfriend and console her and dry her tears!There is an important point implicit here: Suppose from past history of Joe, I know that he is interested in music by Taylor Swift, the recent high end Ford Mustang, the Pats, JavaScript, and vacations on the Gulf of Mexico. Then suppose Joe does a search to find video clips about high performance cars. Well, in that search, I should concentrate just on what he wants in those cars, but for the ads I show during that search I can show him ads for hotels, restaurants, boat charters in the Gulf of Mexico, CDs of Taylor Swift, high performance tires for a Ford Mustang, etc.But that Joe likes Taylor Swift (ah, the girl even if not her music!) will likely mean little or nothing (my math and the corresponding software will decide without my direct involvement) about what hotel he will like on the coast of Texas.So, the point is, knowing quite broadly what Joe likes doesn’t mean much during a particular, narrowly focused search (although some people assume otherwise) but can be quite relevant for showing Joe ads during that search. To be more clear, even if Joe is looking for video clips of high performance cars, someone selling Gulf boat charters still might get some of his business.Oh, by the way, my work is aimed only at the safe for work content, uh, according to, say, the usual US standards. Uh, sorry, Joe!

  4. Chimpwithcans

    Awesome – cant wait to use it. I have made a concerted effort to understand soundcloud better – it is most uncomfortable jump for me to leave the normal music distribution channels for such an open platform, but I am getting there – and features like this will help no end.ps. – typo – 2nd paragraph – ‘their’ = ‘there’

    1. fredwilson


  5. Ana Milicevic

    I’m so excited about this for podcasts – podcast discovery is pretty painful now. I’m sure there’s tons of great content I’m simply not surfacing because it doesn’t hit word-of-mouth channels yet. It’s also a very smart movie by SC – this will turn them from a weekly to a daily must use for me.

    1. sigmaalgebra

      Working on it. Should have all the work done but just need some data on the available podcasts.

    2. Rob Larson

      Agreed. Podcast episode discovery represents a huge hole in today’s listening experience.I don’t want to follow 100 podcasts and manually scan them for interesting episodes. I want an algorithm that will scan everything for me and bring me episodes that it knows I will like.It’s crazy to me that Google or Facebook hasn’t already taken on that challenge. Seems like it would be a trivial exercise for their AI engines. Then insert a 15 second ad in between episodes or charge a premium for ad-free.This soundcloud feature looks like a much-needed step in the right direction, but i would prefer an engine that knows all my preferences to one that just brings me 20 episodes featuring josh kopelman.

  6. Bruce Warila

    “We’ll pull from the 100 million plus tracks on SoundCloud to bring you an endless stream of awesome audio.” If SoundCloud can add “filtered” prior to “track”, I’m in.

  7. Nick Johnson

    This feature looks great. And very important point here: “The difference is that SoundCloud has something like 110 million tracks in it versus something like 2 million tracks in Pandora. That’s important because it means that their are way more tracks to start stations with but even more importantly, the track stations give you access to a long tail of content that you can’t really access any other way.”Agree 100%. This was a big reason why we wrote recently that SoundCloud has more potential than Spotify: http://techcrunch.com/2016/…A lot of power in that network.

  8. Adam Sher

    Music discovery still has a ways to go. On spotify and Pandora, it feels like they aggregate tracks based on external labels as to what type of music it is. I know Pandora explicitly states that this is not the case, but the execution doesn’t belay that intention. Another nit is that across genres that have some similarities, there are too many repeat tracks (typically the most popular tracks that could be counted in the broader genre that overlaps with sub-genres). This shows up especially on Spotify’s radio stations. If SoundCloud better handles this, I’d make the switch.Disclosure: paying user of Spotify and Pandora.

  9. pointsnfigures

    cool. now I just have to find the time to do all this crap……..

    1. Tom Labus

      There’s an app for that!!!

  10. JimHirshfield

    I’d like it to curate a station of laugh tracks for me. Can it do that? I really need that.

    1. Michael Elling


      1. JimHirshfield

        Thanks. I needed that.

  11. JLM

    .The addition of pod casts makes this a publishing endeavor rather than just music.The lesson of Bleacher Report, Bustle.com is this — sample your traffic and double down on the most popular articles in real time.Bleacher Report used this simple idea to take a silo from ESPN and turn it into a company they drove to the pay window. Big time.Bleacher Report was built on the 80 or so most popular sports team which drove 80% of all the traffic. They pitched the deal to ESPN and they showed them the door.While ESPN was tracking everything, Bleacher Report was sampling its traffic in minute frequencies and casting out for new articles on the most powerful traffic streams.In fly fishing they call it “match the hatch.”They harnessed the power of the traffic to tell them where the traffic (revenue) was headed and then when it got there, they had a lot more content for them.This marriage of crowd selected content and skillful, instantaneous analysis/traffic is a winning combination which the same bunch took to Bustle.com.JLMwww.themusingsofthebigredca…

    1. LE

      Why do you think ESPN passed on it? Bad pitch/sales job or complacent execs or?

      1. JLM

        .ESPN wanted to be the most comprehensive such service and prided itself on being able to deliver the data on Slipper Rock State College’s water polo team.These guys wanted to make money.They were young, brash, and right. A very obnoxious combination.JLMwww.themusingsofthebigredca…

        1. LE

          Not a sports guy (for the 100th time) but I can see the appeal of that site.I am waiting for someone to do a site that is essentially the same but focused on the tech industry. In other words complete with not only some stories of substance but also mindless bullshit (I see some of the videos over at bleacher report in other words). Kind of like NY Post or NY Daily news does with news. As opposed to NYT.The site would have, as only one example, paparazzi who follow around and photograph and annoy web celebrities and post their comings and goings. Impossible for me to believe it wouldn’t find an audience. You’d have a portion of the web famous (and there are many as you know) that would be annoyed (like Fred) but many who would actually (just like celebs) welcome the added publicity and mention.

          1. Lawrence Brass

            Something like a tech TMZ? That would be cool!With paparazzis bugging Fred in LA he will probably come back sooner to NYC, where he belongs.

          2. Matt Zagaja

            Valleywag failed.

          3. LE

            Not valleywag I didn’t like valleywag. Execution wasn’t right and not the same concept.Look there are many restaurants that do exactly the same thing some succeed (greatly) and many (maybe even most) fail. Same as with any business. Devil is in the details and the execution. This is something that has been lost in the current day and age that is working hard to insure success instead of just going for the low hanging fruit.You know I’ve told the story of when I started my first business out of college (very early 80’s) and a guy who had been successful in that business (and that knew my Dad) said I would fail because “no tall office buildings nearby won’t work bad location”. But I worked hard and it worked because I found another niche. And I didn’t even know the business at all. Zip. I sold it (after 9 years for a profit, got bored) and it’s still operating today (while my Dad’s friend is out of business but I don’t in all honesty know why that is. ).

  12. sigmaalgebra

    Sounds, not intending a pun, like SoundCloud roughly borrowed Amazon’s “if you like this book (record, can of soup, etc.) then you will also like these other five books (records, …).”Okay. There is some evidence that that technique of discovery, recommendation, etc. can work quite well. E.g., athttp://m.paidcontent.org/ar…isEric Schmidt, Edinburgh Festival Keynotewith Delivering the prestigious MacTaggart Lecture at the MediaGuardian Edinburgh International Television Festival, Google (NSDQ: GOOG) executive chairman Eric Schmidt sought to assure broadcasters: “Google seeks to be your partner, not your foe.” But he warned them: “Listen to the entrepreneurs, not the lawyers, if you want to revitalise your business.” and in the lecture in part We’ve already had a glimpse of the power of recommendations to sway viewing with Netflix (NSDQ: NFLX). Around 60% of Netflix rentals are a result of algorithmically generated recommendations. Another example is Amazon (NSDQ: AMZN). Their recommendations – like “others who bought this also bought” – are incredibly compelling, and in recent years have accounted for between 20 and 30% of their sales. Okay.Notice that there is little to no role for keywords/phrases here. This situation is quite general for Internet content types other than just recorded music and, e.g., as for books, can also be important even for content types based entirely or nearly so on just text.Notice that one of the holy grail problems in computer science is handling meaning of content as appreciated by humans. Well, with the system of SoundCloud, Netflix, and Amazon, say, Amazon, the humans know about the six books and regard them as having similar meanings. Amazon doesn’t know what the heck those meanings are but just that, in the judgment of humans, the meanings are in some significant sense similar.Broadly this is a path into giving users content with the meaning they want.But for something like Netflix, Amazon, and SoundCloud, it should be possible to do better:E.g., with the SoundCloud system, the recommendations are nearly independent of the person using the recommendations and listening. Then, to select what they will like, the listener just listens to a long list of recorded musical pieces and selects the ones they want. This is a little like looking for a book on computer science by going to the computer science collection in a library and looking at all the books one by one sequentially.So, there should be a better way.There is much more ….

    1. LE

      Long before Amazon I always wondered why, when I went into a video rental store, they didn’t have display racks with, as only one example, “You like ‘the Godfather’ you will also like these movies as well”. Common sense.The reason is the same reason that Nicolas Dinale’s comment above said “I wonder what SoundCloud is doing to pull long-term Pandora users over”. Many web services don’t do much of what I would call traditional marketing or even guerilla marketing at all.When video stores were popular (and before they starting going out of business) they didn’t need to really break a sweat to get users. It was a unique product and they could just offer it and the world would flock to it effortlessly if they had a good location. So they got lazy. They never developed the hungry skills needed to actually grab customers because they didn’t really need to. Wouldn’t have been that difficult to put up a few displays and arrange movies that way. Maybe certain stores did that but I was in a great deal of video stores and I never saw it (at Blockbuster as one example).Ditto for wine stores. I can go into a liquor store and get confronted by a supermarket of liquor. Never see liquor arranged by “you like Riesling try these gewurztraminers”. [1][1] The only reason I know of those two wines is that my brother in law is into wine I’d still be stuck on zinfandel if not for him.

      1. sigmaalgebra

        Sure, darned near every retailer needs a system for customers to use for search, discovery, recommendation, curation, notification.Sure, for Wal-Mart selling DVDs, need such a system plus, of course, a way to find DVDs better than searching through a huge bin of a few hundred just tossed in there.The math I derived and the code I’ve written should be useful for such cases, that is, beyond just a Web site. But, sure, don’t necessarily have to work much or at all with Wal-Mart since customers can just use the Web browser on a smart phone while in the store.Right: Once a week have someone go through Sam’s Club and find all the items they are selling and where they are located in the store. Do this for each of the stores. Then have a Web site that has this info and lets a user, in any particular Sam’s Club, do a search for what they want. Sure, then for the whole shopping list, solve the traveling salesman problem, that is, find the shortest path in the store that touches each product to buy just once. Then, if do this for Sam’s, then Wal-Mart, A&P, Costco, etc. may be willing just provide the data. Get a little barrier to entry from a network effect.”zinfandel” — that one’s weird because no one in the wine business knows where the heck it came from. Maybe now with DNA analysis could say, but grapes are weird because they don’t “reproduce true to seed”, that is, plant a grape and can get nearly anything, not much like its parents. That’s why, to grow, say, Pinot Noir, can’t just get a grape and, instead, have to have a cutting of a stalk.But, sure, Chardonnay: It’s the grape of the best white wines of Burgundy, e.g., in and around Macon. So, that covers also Hitchcock’s favorite Montrachet, now absurdly expensive. But nearly any reasonably goodappellation d’origine contrôlée near Macon to my taste is a lot like the really expensive stuff or the moderately expensive stuff, e.g., Pouilly-Fuissé.Similarly for Pinot Noir and the red wines of Burgundy so that Nuit St. George is darned near just the same as Corton or Chambertin, etc.

      2. sigmaalgebra

        > GodfatherYes, my project should be good also for movie recommendations and, there, really should do much better than the Netflix recommendation system. Netflix posed the problem poorly, so poorly that a good answer is impossible.

  13. Nicolas Dinatale

    I’ve spent years curating my paid Pandora experience. I wonder what SoundCloud is doing to pull long-term Pandora users over?

    1. Michael Elling

      Other than with native content here?

  14. Dorian Benkoil

    I’ve always considered SoundCloud something of a “sleeper.” It doesn’t get the big buzz of some others, but has built consistently over the years. Each of these features — including the one you just noted — helps and is an improvement. Here are a few on my wish list as a media producer and strategist: 1. Better ability for artists posting to monetize their work in an easy rev-share model, as is done with YouTube or Apple or Amazon (for the record, the artist should get 70%, not the 55% of YouTube). Make it super easy and take a cut. Win-win. Even better would be added ease of e-commerce, paid downloads and even ability to easily “paywall”, give more access for a fee/subscription/etc. Could pay off for business-oriented audio, too. Something I haven’t seen. 2. Will this (as is alluded below) be good discovery for spoken word — podcasts, comedy, etc.? There is not a great “find more like this” for podcasts or comedy that I have seen. It’s a niche, but could pay off over time once created. This can be short- and long-form. Spoken audio can be much shorter — modularized if you like — from the roughly half-hour to hour model prevalent now. Just as people can see their favorite clips from an hour-long show such as Jimmy Fallon’s on video platforms, why not have same for audio? Again, win-win. // I’m convinced that there are a lot of people who would like to consume audio beyond music with the same facility and ease of video in their cars, while walking, etc. Let’s let them have it.

    1. Ryan Frew

      The biggest thing SC is missing (and I use it religiously), is consistent posts by big artists. If I want to add radio songs to my playlist, it either won’t be there, have been pirated and added by another user, and/or buried under a pile of cover songs.

  15. LE

    1) I’d like to see if they can figure out a way to also take “little bits and pieces” of audio content (over the air broadcasts or podcasts ) and string them together in the same way. So you are not trapped in one piece of content but can be exposed to many different things (sort of like an “audio trailer” I guess). Then with an easy way to reference what you just heard so you can find the full content if you want.For example a few of the podcasts that you mentioned seemed interesting but I don’t want to take the time to listen and find out by actually playing the podcast.2) Noting also that the soundcloud homepage says “find the music you love. discover new tracks. connect directly with your favorite artists”. ie doesn’t mention podcast discover anywhere on the web home page. How would I know I could use it for that? (I am not a soundcloud user because I am not into music..)…

  16. creative group

    FRED:Just the reference to track had us reflect upon the crane that collapsed and is diverting the work commute this morning in Tribeca. It is unfortunate one died but fortunately no other deaths occurred.

  17. creative group

    FRED:If your company didn’t have a financial stake in many of the company’s you promote it would be unlikely you could use everything and actually be productive during a work day.Based upon the production required to be and maintain a level of proficiency and success in any occupation. IOHO.

  18. creative group

    Soundcloud for podcasts are useful. We are trying to figure out what we used before it. The podcasts just had loaded automatically with NPR or NBR.

  19. creative group

    Maurice White the co-founder of Earth, Wind and Fire died today after battling Parkinson’s disease. (They say it occurs in threes, Frey, White, Bowie) Three great musicians.https://youtu.be/_Vbo6K00ZPAThree hits our older contributors may remember.Shining Starhttps://youtu.be/rl-WSmryfSYDevotion (Couldn’t locate the live performance. Older live shows are golden)https://youtu.be/97PompF6C44Can’t hide lovehttps://youtu.be/wr0ekZlcM8EEnjoy!Earth, Wind and Fire on Soundcloud all weekend…

    1. Lawrence Brass

      Luckily for us their music transcends. Devotion specially, from the first notes, has the power to transport me to wonderful moments of the past. Thanks Creative.

      1. creative group

        Lawrence Bass:Your are welcome.Great music bonds and brings the best together..

  20. William Mougayar

    I’ve always liked the sideways swipe for discovery on SoundCloud. It’s one of those hidden features

  21. Joy Kennelly

    Fred, Spotify does the first part, but not the podcast aspect which makes me want to check it out. Also, just checked and Spotify only has 2 million tracks which was interesting to learn since it’s normally my go-to music source. Now that they’re playing a commercial after every song I listen to (since I refuse to pay to upgrade:) I’ll check out your site instead. I never got into Pandora because it was too mainstream and I prefer to hear new music. The only thing with Soundcloud was so many indie artists I discovered just weren’t my taste. Perhaps I’ll find something new now though. And the podcasts sound interesting too. Thanks!

    1. Tyler

      Soundcloud is the best place to find new music. I rarely use Pandora for the reason you listed – you end up listening to the same tracks over and over and over and over…..

  22. Ryan Frew

    Every song in the Soundcloud web app includes an area on the right side titled “Related Tracks”. It shows 3 songs, and you can click on “View All” to display an unlimited number of similar songs. It will play them automatically, just like a Playlist. Isn’t this the same functionality as Track Stations?

  23. Lawrence Brass

    I think I am still a Grooveshark widow, however I have to recognise that Soundcloud has owned my workstation speakers lately. Maybe its time to try it on the devices.

  24. sigmaalgebra

    Gee, this just in:Claim: Zip, zilch, and zero competition for anything like AI. ML, etc. or any of its hoped for applications from Google!Proof: Look at the backgrounds of their people, e.g., athttp://www.businessinsider….QED!I couldn’t have picked better people to compete with!

    1. Ellie Kesselman

      I just wandered on over from Big Red Car’s blog. I saw a post dissecting the hype from AI and ML, referencing some of the names in that Business Insider article in particular that I think hope you’ll enjoy: “Big Data is Dead. All Aboard the AI Hype Train!” https://medium.com/@adailyv

      1. sigmaalgebra

        Yup. Thanks for the link.Yes, the points in the link are quite good.Yes, what your link said about IBM, etc. pushing AI is really just hype. At least they want publicity. For more they think they have something they can sell and are trying to sell it.For more on hype, there ishttps://news.ycombinator.co…That post talks about the next big thing, System-K.While the article in your link did pull some of the hot air out of big data, artificial intelligence, and machine learning, it didn’t explain a good alternative.Here is one: We gather data, use computer hard/software to manipulate it, and report results. We want the results to be valuable, and from history sometimes they are.Well:(1) Essentially, we want the results to be valuable for some question we have. So, when we pick the input data and the data manipulations, we must have the question in mind, that is, we need data and manipulations good for the question. Simple.But so far essentially always, we need to have a fairly specific question in mind. That we can collect some data and write some software and, then, get valuable answers to a wide variety of questions is essentially delusional.(2) Much of the potential value is from how we manipulate the data. There are various ways:(A) Manual. Maybe the data manipulations we want the computer to do are ones that we have done by hand or at least in principle could be done by hand. So, we program those manual manipulations.(B) Intuitive Heuristics. We make some guesses about what manipulations might be effective and program those. That’s, say, using the TIFO method — try it and find out. Usually we have no idea if we have results close to the best possible….(C) Math. The manipulations are, at least in principle, necessarily mathematically something.So, for more valuable results, maybe proceed mathematically. Here I mean with theorems and proofs. Why? Because the results of such work are essentially the highest quality, and often most powerful, knowledge in our civilization. For how theorems and proofs work, a good high school course in plane geometry is a good example. Calculus is a better example when the course states all the important results as theorems and proves them, e.g., as in W. Rudin, Principles of Mathematical Analysis.For applications, such as we are considering here, the math is a little like an airline serving many small islands in the South Pacific. So, if by boat or whatever you can get to one of those islands, then the airline can get you to any of the other islands. Then, maybe with another boat trip, you can get to your real destination.But, to begin you need that boat trip to the first island — that’s like assumptions. Then each airline hop from one island to another is like a theorem that starts with assumptions and yields conclusions. At the final conclusions, you take another boat trip to your final destination, the results you need.Better math, a better airline, can get you to better places!Example 1: Out in the flat, dry lands of Texas, you want to get rich by drilling for oil. But where?Well, can set off a minor explosion and send a pulse of sound waves through the ground. As is the case commonly with waves, for each wave, when it goes from one medium to another, say, from limestone to granite or maybe from granite to sand with oil, some of the wave gets transmitted and the rest, reflected.Underground, there are lots of layers that generate lots of reflections. Some of the waves get reflected several times.So, on the surface can have some microphones that receive the reflected sound waves. Record those.Now would like a 3D map of the underground layers. The shapes of the layers can give valuable clues about where to drill.So, how to find the 3D map of the layers from the reflections? Well, if take a wave, delay it, and add it to itself, then that is a convolution. The convolution is still a linear operator on the wave. And if take the Fourier transform of a convolution, then with meager assumptions, get the same thing as multiplying the Fourier transforms of the waves before they are added. So, with some use of Fourier theory, can do a deconvolution and, thus, find each wave from each reflection and, thus, each underground layer.Details are in, say,Enders A. Robinson, Multichannel Time Series Analysis with Digital Computer Programs.But the computing needs to do a lot of Fourier arithmetic.One day R. Garwin of IBM Research was at a President’s National Science Advisory Committee meeting sitting next to J. Tukey, of Princeton and Bell Labs, where Tukey was taking meeting notes with one hand and doing Fourier derivations with the other. Garwin was having trouble making Fourier arithmetic fast enough and asked Tukey. Back at the lab, Garwin had J. Cooley program what Tukey said, and the result was the Cooley-Tukey fast Fourier transform. It revolutionized a lot in signal processing, including for the oil patch and bought me a high end Camaro hot rod!So, the math was some Fourier theory. The assumptions were from the physics of waves their their reflections. The results were 3D maps of the underground layers, and commonly those maps were very valuable.There is a super nice, precise, succinct presentation of Fourier series in Rudin above and of Fourier transforms in W. Rudin, Real and Complex Analysis.That’s a good example of how to do it, that is, how to start with a question, take data, manipulate it using some math, and get valuable results.It’s applied math, sometimes with some original math, and the technique, paradigm, whatever, is rock solid, mostly not nearly new, and often very valuable.And, like anything good, the technique has nothing, zip, zilch, and zero, to do with anything like what is now called artificial intelligence, machine learning, big data, or any of the rest of that nearly totally worthless, incompetent, deceptive, delusional, misguided, ignorant, totally upchuck-able nonsense.That computers are “giant electronic human brains” has been part of IBM marketing hype going way back. We are hardly any closer now than then.To be clear:(1) The applied math technique I described has many thousands of great, very valuable examples, with rock solid logic, going way back to Newton and Euclid.(2) In outrageously strong contrast, AI, etc. have no great examples, in total only a few examples, with nothing like comparable value, and with the logic on the level of tea leaf reading.E.g., for that example I gave of a use of Fourier theory, the logic is clear and solid. But for the deep learning, convolutional neural networks, the associated workers admit that they have little understanding of how it works when it appears to, e.g., recognizing faces of cats.Or their paper might read: “We have some software that appears to recognize images of cats, but we don’t know how it works.”.For more, in AI, take away the use of classic statistics and don’t have much left of any value.With AI, etc., we’re rarely talking anything “new, correct, and significant” and, instead, are usually talking adulterated classic statistics with new labels.Here is one of the mistakes: Since often we need computers to do the data manipulations and arithmetic, we start to assume that what data manipulations and arithmetic we should do should come from the field of computer science. Wrong. That’s like saying that the experts in making wedding cakes are the guys who grow wheat, sugarcane, and eggs and make cream and butter. Those guys a lot more about electric motors and roller bearings than genoise, ganache, buttercream, biscuit au beurre, Sacher Torte, etc.Ah, when you are in a hospital for a serious problem, you get your care from the best qualified physician you can find and not from the guy who cleans bedpans.