Does the Vocal Removal Work

[oluwabajio]

I tested it on my phone but it wasnt working as expected. Were you able to get spleeter to work effectively.

[farihachaiti]

Yes. You have to update the ffmpeg command as per your requirement. This is the base of android implementation of spleeter only. You have to do the minor tweaks yourself.

[oluwabajio]

Yes, you are correct, i made a lot of tweaks and was able to get it up, but it crash at the ffmpeg command. I would try again today and get back to you. Thanks.

[oluwabajio]

Ok, i tried again. This is the error i am having.

java.lang.UnsatisfiedLinkError: dlopen failed: library "libmobileffmpeg_abidetect.so" not found

[oluwabajio]

Ok finally, i was able to get it to work on my device. Thanks for your help.

Also i noticed the result karaoke was of very low quality. Is there any way the quality can be increased

[oluwabajio]

i went through your code and later discovered that you are actually not using spleeter to extract the karaoke. You are basically using ffmpeg to perform phase cancellation on a stereo track.

That explains the reason for the low quality audio.

Nice work by the way.

[farihachaiti]

@oluwabajio Check correctly. I am only using ml tensorflow model here which enables the ffmpeg to split the vocal. This is the best possible result till now as an android solution of spleeter. You are free to extend it and make it better. You can also use the denoising processes to make the audio quality better which I commented out in this implementation. Good luck.

[oluwabajio]

Ok, thanks for your response. From your code, i noticed that after you used tflite model to split the audio, you were supposed to encode the audio data to an output format, but instead of executing the encoding command, you instead executed the command for phase cancellation to generate the output audio. phase cancellation = String command = "-i " + audioInputFile + " -af pan=\"stereo|c0=c0|c1=-1*c1\" -ac 2 " + audioOutputFile + " -y"; Encoding = String ffmpegCommand = "-f f32le -ac 2 -ar 44100 -i " + pipe1 + " -b:a 128k -ar 44100 -strict -2 " + outputPath_1 + " -y";

Also checking through vvasanth project, This was the command he used. val ffmpegCommand = "-f f32le -ac 2 -ar 44100 -i " + pipe1 + " -b:a 128k -ar 44100 -strict -2 " + outputPath_1 + " -y"

![Uploading command3.png…]()

[farihachaiti]

@oluwabajio Oh yeah. The encoding command is the ffmpegCommand. You will see there I had tried multiple commands. I was just playing with the various commands to check the ffmpeg running or not. This project was in python basically. You are free to work with this android solution. Please make the ffmpegcommand work, cut my commentouts and do play as you need. 👍

[oluwabajio]

ok cool. Thanks for your feedback. I would try the encoding format and get back to you.

[farihachaiti]

Hey @oluwabajio This file is the working version of Spleeter, I forgot to update last time. Btw this throws out of memory error in my device with incomplete music output along with some noise, so please check on your device if works correctly. I guess if it could work completely without throwing any memory error the output would have been perfect. Moreover, I suggest you play with the ffmpeg commands and the denoising process to get the desired output if it still troubles. https://github.com/farihachaiti/DoKaraokeAI/blob/main/app/src/main/java/com/example/dokaraokeAI/SplitVocal.java https://github.com/farihachaiti/DoKaraokeAI/blob/main/app/src/main/java/com/example/dokaraokeAI/AudioFeatureExtraction.java

[oluwabajio]

ok cool. i would check it out

[oluwabajio]

ShowProgress.java Class is missing.

[farihachaiti]

@oluwabajio Yeah I intentionally didnt put it in. Its a progressdialog only. Please comment out these lines linked to showprogress.java. Run the application without the progressdialog.

[oluwabajio]

@farihachaiti i checked it out. it works fine, but the output result was really not good. I want to find out, is there ay other way we could increase the quality of the audio apart from the denoising method.

[farihachaiti]

@oluwabajio sure. Go ahead. 👍 If you find anything please open another issue. I am closing this one.