PLEASE REGISTER TO POST. Also, be sure to visit the main website www.tedgreene.com

**************************************************************************************
Buy NOW on Amazon
My Life with The Chord Chemist
A Memoir of Ted Greene, Apotheosis of Solo Guitar
Available at amazon.com

*Check it out!!!

VISIT OUR NEW SUPPORT PORTAL
Your contributions keep the site healthy and growing


More information HERE

Official Ted Greene Archives Blog

Ted Greene Archives on YouTube

Join Ted on FACEBOOK

NEW! Follow on TWITTER

..:: The Ted Greene Forums ::..
Sign up Latest Topics
 
 
 


Reply
  Author   Comment  
xavriley

Registered:
Posts: 5
Reply with quote  #1 

Hi everyone,

I shared this in the Ted Appreciation group on Facebook, but I'm pleased to say that I've found a method for reducing the background noise from the "Live at the Seashell" videos on YouTube. There's a playlist of the remastered versions here:

https://www.youtube.com/playlist?list=PLHyebay_8doeQno0fMFIoM1Zs-YJRuT41 (contains all 9 videos)

A few points - this isn't just noise reduction or EQ. Instead it uses some new AI technology to separate the guitar from the conversation "intelligently" and I think the results are pretty good. The tone of the guitar is still there and clearer in some cases.

I did play around with other remastering techniques and EQ but it didn't make a huge difference - I'm happy to provide the extracted audio files to anyone if they wanted to have a try themselves.

Also, this is only based on the videos that I pulled from YouTube - they're over 10 years old so the resolution is quite low. If anyone here could provide me with a higher quality transfer of the original footage I'm confident that we could produce something even better.

Finally, I'd love to know if any more footage exists - even if the audio was previously considered unusable, this new process has been quite impressive and might yield better results.

I hope everyone enjoys these - I'm glad to be able to give something back to this community and honour Ted in whatever way I can.

Michel

Registered:
Posts: 85
Reply with quote  #2 
Super work, thank you. [thumb]
__________________
Nothing but music.
PaulV

Moderator
Registered:
Posts: 1,754
Reply with quote  #3 
Can you please tell me what program was used?
Thank you.

__________________
--Paul
xavriley

Registered:
Posts: 5
Reply with quote  #4 

Hi Paul,

Sure - it's a tool called Spleeter which is made by Deezer (an online music streaming company). The easiest way to use it by far is an online service called ezstems: https://ezstems.com/

To explain what's going on, the technique is known as "source separation" and there are several programs that can do it. It's an active area of research in the MIR (music information retrieval) field. Until recently it was extremely expensive and laborious and the results weren't that impressive. In the past couple of years there have been several breakthroughs with machine learning that make it more practical now.

The specific settings I used for spleeter were the 2stem model at 11kHz. There are also options for analyzing frequencies up to either 11kHz or 16kHz but they didn't make a difference in my assessments (could be my ears though...).

To actually run the program I used a tool called Docker which is a way of running Linux based programs on any machine. The command was as follows:

$ docker run -v $(pwd)/output:/output -v $(pwd)/input:/input -v $(pwd)/models:/model -e MODEL_PATH=/model researchdeezer/spleeter separate --verbose -i ./input/ted_seashell_4.wav -o /output -p spleeter:2stems-16kHz

The way this process works does consume a lot of memory though - I had to give Docker 10Gb of ram in order to finish the processing for one segment. If you need to process long files I recommend splitting it into chunks first if possible.

Spleeter isn't actually designed to remove conversation as such, but it just happened to work well in this case. The models they have are trained to separate a song into separate instrument tracks (aka "unmixing") so that they can perform things like pitch or beat analysis at a large scale. The model you use determines how many stems are output - the 2 stem model outputs "vocals" and "accompaniment", the 4 stem model outputs "drums", "bass", "vocals" and "other" and so on. The results are pretty mind blowing on standard rock and pop tracks. I put some Jonny Smith through it too and it managed to isolate the guitar pretty well.

I realise all this is very technical but I'm happy to answer questions if anyone has a project in mind.

I'm also still interested in producing a higher quality version of the seashell recordings if anyone can transfer a copy to me. I'm happy to pay for it - I just don't know where they are available from.


Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.

YOUR SUPPORT MAKES A DIFFERENCE :: DONATE