Identifying radio streams

[Eduardo Avaria]

Hi guys, 

I developed a software on my own (based on this paper) a few years ago to fingerprint the radio stream every 3 seconds and query a database of select audio pieces previously digested (individual commercials) to check when they were getting aired. If some piece was not recognized, a human operator could come and add it to the DB, making future transmissions identifiable. After a few months of trying to sell my prototype with no success, I abandoned the project.

Now I'm starting to work on it again and would love to know the opinion of someone who knows about echoprint. 

Do you think that with at most *some* tweaking to the code could be possible to reproduce such functionality, Specially the need to identify 30 seconds spots (with the possibility of a sample including 2 different spots). I believe that using 10s samples should solve this, but I'm not sure if can be done with enough accuracy, considering that some spots don't even have music on them.

Should I stick to my code and if there's no match there, query echonest trying to find a song, or is it worth it trying to move the whole system to echoprint?

Hope you can give me some insight. 

Regards,

[Henk Beld]

Hi,

I'm looking for a solution you describe and have been looking into using echoprint. Would like to discuss the options with you.

Cheers,

Henk 

Op maandag 26 mei 2014 03:23:55 UTC+2 schreef Eduardo Avaria:

[Christo Grozev]

Dear Eduardo,
I am currently writing a system that appears to be identical to what you already wrote a few years back, as you say. Any chance you can share some of your approach to implementing the algorithm described in the Phillips article?

thanks

Christo

[Tony Li]

Hi Henk,

Doreso(http://doreso.com) provides a radio music monitoring solution. Please feel free to let me know if you are interested.

Thanks

Tony

[henrique matias]

Hello Guys / Tony i would be very interested in identifying tracks being steamed via "icecast" compatible stream.

Any ideas on how to proceed?

Would be good to work with a partner that would be always updating the database so we could offload this fingerprint task to some specialists.

Please feel free to get in touch,

peace

[Chima Kaluorji]

Hi Eduardo Avaria,
I have been researching for weeks on doing similar thing you did. What will it cost to get the software and source code of your project explain below. I am a master degree student, repeating what you did is what I suppose to do in my project.

Thanks

On Monday, May 26, 2014 at 3:23:55 AM UTC+2, Eduardo Avaria wrote:

[Eduardo Avaria]

Tell me a little more about your project and whatsapp you can oferta and i can share me cose. The project is currently on pause, but i plan to finísimo it dome time soon. I can even consider mami y it open source if there's interest.

--
You received this message because you are subscribed to a topic in the Google Groups "echoprint" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/echoprint/TqCRFRQe6vU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to echoprint+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Chima Kaluorji]

Actually I am a Master Degree Student and audio processing is my project topic, although I work with Ebonyi Broadcasting Corporation (EBBC), so I am using my working place as my case study.

Project
I bought a software (VRS by NCH) that enable me to record all the daily program in an interval of 4 hours.

I want to track all the advert played within the 4 hours recorded audio file.

On the GUI, I will have three buttons,

The first button will enable me to select the 4 hours recorded audio file. Process it with FFT and store the audio fingerprint in the database.

The second button will enable me to select a folder containing all the advert files, then I will like to loop through all the advert files and store the fingerprint of each of the files in the database. (We can select the file one by one if it will take time and memory space to loop through all the advert).

The third button will enable me to use the advert audio fingerprint to cross check if it is contain in the recorded audio fingerprint. If it is true then send the name of the advert file, duration of the advert file, date of cross matching the two audio files (recorded and advert) and date of recording the recorded file which normally comes with the name of the recorded file (i.e. 2015-09-17 07-26-01 Channel 1).

Finally, We will have a print button that will enable me print the name of the advert file, duration of the advert file and date of recording the recorded file which normally comes with the name of the recorded file (i.e. 2015-09-17 07-26-01 Channel 1) using date of cross matching the two audio files (recorded and advert) to fetch the above data.

In summary, it is an advert tracking software.

To unsubscribe from this group and all its topics, send an email to echoprint+unsubscribe@googlegroups.com.

[Eduardo Avaria]

Hi, I have created the exact functionality that you describe.

The problem with that approach is that id doesn't scale with data size, and you are not addressing the problem of matching the fingerprints (it's not an exact match). Anyways, that problems are not an issue for the project that you describe.

In short, I have more or less the following software pieces:

A Java library that fingerprints any audio file in the correct format and manages writing fingerprints files.

An interface that converts any audio file to the correct format using FFMPEG

An Arduino based FM recorder that feeds a Hadoop server using Apache Flume

An internet stream recorder (opensource) that will also feed the Hadoop using Flume

A small program that uses the GraceNote api to identify songs.

And a bunch of Pig Latin scripts that using fingerprint text files (from the stream and the spots) with some intelligence makes the reports you are describing on large amounts of data. These reports are meant to be official documents to certify the transmissions, interactive control panel with charts and statistics and semi structured data to develop some intelligence on them.

The last feature I was developing was using deep neural networks to try to identify specific words (like the name of a politician) on the stream as well as identifying the person speaking by his voice features. Sadly, training the network requires a lot of labeled training data that I'm not able to get... there's a little development on that side, but it's worth mentioning.   

There's a lot of room for improvement (a lot of experimentation has been done and everything is more or less prototypes), and right now I'm not developing it anymore since I never really got a client willing to pay for it, so I'm open to suggestions on how we ca proceed.

I considered making the project Open Source, but there is a substantial effort to gather and document everything in a way that can be useful, maybe you can adapt your project to include the release.

Anyways, tell me what is your idea and we can work on it. Maybe even if EBBC is willing to make an offer, I can give you whatever you need, including the support to develop any missing piece you may need.

I'm looking forward to make something about this.

Regards

Eduardo.

To unsubscribe from this group and all its topics, send an email to echoprint+...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

You received this message because you are subscribed to a topic in the Google Groups "echoprint" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/echoprint/TqCRFRQe6vU/unsubscribe.

To unsubscribe from this group and all its topics, send an email to echoprint+...@googlegroups.com.

[Chima Kaluorji]

Wow, this is a great developing. I have been searching for help for days.

EBBC will not be interested in the cross-matching application, because they feel that VRS recorder from NCH is doing the job for them. But I am 100% sure they will be happy to have a software that can stream television and radio signals online (on there website). They have even mentioned it to me.

But the one that is giving me sleepless night is the cross-matching application, I suppose to submit it to my supervisor today, and yet nothing is happening, although I have called him to ask for extra time.

So what we will do is that, I will pay you per hour to help me complete the cross matching application, may be, we will use TeamViewer so that I can see what you are doing and learn while you are doing it. Then, let me know how much you will charge for the tv and radio online stream application so that I can prepare a proposal for the Managing Director.

Finally, you have to be moderate on your billing because, in this country, $5000 is already a million Naira (Nigeria) in local currency.

In summary, I will like us to start today because, as I said before, I am already behind schedule. I will like to know the method of payment too. We can communicate using my email address: chimak...@gmail.com so that I can get your respond quickly on my phone.

From my research, they use hash to store fingerprint into the database and humming to search for comparison, may be, this information may be helpful.

Thanks, Hoping to hear from you soon.

To unsubscribe from this group and all its topics, send an email to echoprint+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.