Update on Echoprint development

[Andrew Nesbit]

Hello Echoprinters,

I have recently started working with The Echo Nest to help take care
of Echoprint. I am writing to let you know about the current status of
Echoprint and to give you an idea of our plans for its future
development.

Over the past several months, significant changes have been made to
Echoprint's server-side matching algorithm. Although these changes
were beneficial in many respects, one unintended consequence is that
many queries which used to work with The Echo Nest's Echoprint server
no longer work. There have been several reports via this group that
queries which are expected to return positive matches, such as the old
examples on the http://echoprint.me/ website, actually return no
results at all. We can confirm that this is a problem, and we agree
that it is confusing. We have a good understanding of what is causing
it, and we are already working on a fix to the matching algorithm.
Once it has been tested properly we will be able to install it on the
production server.

Other things we are working on include testing over-the-air (OTA)
matching performance and noise-robustness, and resolving reports of
various bugs.

Rest assured that we are actively looking into the issues which have
been raised during the past couple of months. For the meantime,
however, I have updated the examples at http://echoprint.me/ so that
the Echo Nest's Echoprint server returns correct matches for the
current version of the matching algorithm. There are more changes
forthcoming to both Echoprint and its website, but the information
posted on the website should now be correct. Please let me know if you
find something that still does not work as you would expect.

Finally, please note that ENMFP is not affected by anything I've
described here, as that is a different service.

Thanks for hanging in there whilst we update things on our end, and
for your patience. I will post another update as soon as we have more
information.

Andrew

[Seppo]

Thanks Andrew for the heads-up. I was struggling quite a lot as this
project sound so promising but in the end I couldn't get many pieces
to work.

What I did was first do a Android test using the phone's built-in
microphone. First I tried the public server and as it really didn't
return any results I put up my own server.

The own server didn't really work with OTA data either. I tried it
with my own Android implementation and with Alex Restrepo's project.
Neither returned any results.

I also recorded with a real microphone and ran the resulting MP3
through EchoPrint and got no results whatsoever although the sound
quality sounded really good.

Any "real" MP3 was matched perfectly by EchoPrint.

So I guess at this point I have three questions:
1) Can you offer a rough ETA for the server fixes?
2) What can you say about the current/future OTA match performance? I
haven't really had any luck with OTA so far. It's vital for my
application though...
3) What about the 7 million new songs that you talked about last
summer, will they be eventually included in the public server?

Kind regards and many thanks,
Seppo

On Feb 7, 10:35 pm, Andrew Nesbit <anes...@echonest.com> wrote:
> Hello Echoprinters,
>
> I have recently started working with The Echo Nest to help take care
> of Echoprint. I am writing to let you know about the current status of
> Echoprint and to give you an idea of our plans for its future
> development.
>
> Over the past several months, significant changes have been made to
> Echoprint's server-side matching algorithm. Although these changes
> were beneficial in many respects, one unintended consequence is that
> many queries which used to work with The Echo Nest's Echoprint server
> no longer work. There have been several reports via this group that
> queries which are expected to return positive matches, such as the old

> examples on thehttp://echoprint.me/website, actually return no

> results at all. We can confirm that this is a problem, and we agree
> that it is confusing. We have a good understanding of what is causing
> it, and we are already working on a fix to the matching algorithm.
> Once it has been tested properly we will be able to install it on the
> production server.
>
> Other things we are working on include testing over-the-air (OTA)
> matching performance and noise-robustness, and resolving reports of
> various bugs.
>
> Rest assured that we are actively looking into the issues which have
> been raised during the past couple of months. For the meantime,

> however, I have updated the examples athttp://echoprint.me/so that

[Andrew Nesbit]

On Thu, Feb 9, 2012 at 1:38 PM, Seppo <se...@comeks.com> wrote:

 

What I did was first do a Android test using the phone's built-in
microphone. First I tried the public server and as it really didn't
return any results I put up my own server.

The own server didn't really work with OTA data either. I tried it
with my own Android implementation and with Alex Restrepo's project.
Neither returned any results.

Hi Seppo. Thanks for the message. I am not sure what the cause of this would be, but I would imagine it is related to one or both of two things:

1. The need for changes to the matching algorithm I alluded to in my earlier message. This will affect results returned by The Echo Nest's main Echoprint server and also your own server if you update it after we release the changes to the community. I would hope that these changes improve OTA performance, and we would need to test for this specifically. Our first priority is to fix a few immediate bugs, then we will continue to work on OTA performance.

2. The need to improve OTA performance more generally. This also entails changes to the matching algorithm, but it may affect other parts of the design of Echoprint too.

I also recorded with a real microphone and ran the resulting MP3
through EchoPrint and got no results whatsoever although the sound
quality sounded really good.

Generally speaking, this sort of thing is one of the biggest challenges in audio signal processing. It is a reasonable expectation and a simple concept that if two pieces of music sound the same to a human listener, then they should return almost identical audio fingerprints. Echoprint does well at this, particularly at the file level, but we want to make it even better, especially for OTA matching such as you have described.

Any "real" MP3 was matched perfectly by EchoPrint.

Good to hear. This is what we would ordinarily expect.
 

So I guess at this point I have three questions:
1) Can you offer a rough ETA for the server fixes?

My estimate is that this will take a couple of weeks. In order to minimise confusion regarding changing implementations, it will need to undergo testing against some large data sets before we can put it into production.
 

2) What can you say about the current/future OTA match performance? I
haven't really had any luck with OTA so far. It's vital for my
application though...

I cannot make predictions about future OTA performance until I have got the right tests underway. This is high on our list of priorities.

3) What about the 7 million new songs that you talked about last
summer, will they be eventually included in the public server?

We are now working with other partners and discussing this very issue to see what we are allowed to make publicly available.

Andrew

[Seppo]

Great, I'll eagerly wait for any updates.

Thanks again.

Seppo