Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Deep neural networks for separating singing voice from music written in TensorFlow
| Date | Stars |
|---|---|
| 2026-07-31 | 800 |
| 2026-08-01 | 800 |
| 2026-08-02 | 800 |
| 2026-08-06 | 800 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Deep Neural Network for Music Source Separation in Tensorflow >__This work is from [Jeju Machine Learning Camp 2017](http://mlcampjeju.kakao.com)__ >* Co-author: Mark Kwon ([email protected]) >* Final work will be done in Jeju ML Camp. Please check [here](https://github.com/hjkwon0609/source_separation_ml_jeju). >* [__Take a look at the demo!__](https://www.youtube.com/watch?v=Cx7Me0Ayz1I&feature=youtu.be) ## Intro Recently, deep neural networks have been used in numerous fields and improved quality of many tasks in the fields. Applying deep neural nets to MIR(Music Information Retrieval) tasks also provided us quantum performance improvement. Music source separation is a kind of task for separating voice from music such as pop music. In this project, I implement a deep neural network model for music source separation in Tensorflow. ## Implementations * I used Posen's deep recurrent neural network(RNN) model [2, 3]. * 3 RNN layers + 2 dense layer + 2 time-frequency masking layer * I used iKala dataset introduced by [1] and MIR-1K dataset which is public together when training. ## Requirements * Numpy >= 1.3.0 * TensorFlow == 1.2 * librosa == 0.5.1 ## Usage * Configuration * config.py: set dataset path appropriately. * Training * ```python train.py``` * check the loss graph in Tensorboard. * Evaluation * ``` python eval.py``` * check the result in Tensorboard (audio tab). # \[Related Paper\] Singing-Voice Separation From Monaural Recordings Using Deep Recurrent Neural Networks (2014) <sup>\[3\]</sup> ## Proposed Methods ### Overall process * Waveform of a music(the mixture of voice and background music) is transformed to magnitude and phase spectra by Short-Time Fourier Transformation(STFT). * Only magnitude spectra are processed as input of the RNN layer. * Estimated magnitude spectra of each sources and phase spectra of the mixture are transformed to waveform of each sources by ISTFT(inverse STFT). <p align="center"><img src="https://raw.githubusercontent.com/andabi/music-source-separation/master/materials/posen/overall.png" width="75%"></p> ### Model * RNN layers (3 layers) * Dense layer * 1 for each source * Time-frequency masking layer (normalization) * 1 for each source * no non-linearity * src1's magnitude + src2's magnitude = input's magnitude <p align="center"><img src="https://raw.githubusercontent.com/andabi/music-source-separation/master/materials/posen/model.png" width="75%"></p> ### Loss * Mean squared error(MSE) or KL divergence between estimated magnitude and ground true are used as the loss function. <p align="center"><img src="https://raw.githubusercontent.com/andabi/music-source-separation/master/materials/posen/mse.png" height="30px"></p> <p align="center"><img src="https://raw.githubusercontent.com/andabi/music-source-separation/master/materials/posen/kl.png" height="30px"></p> * Further, to prevent different sources to get similar each other, 'discrimination' term is considered additionally. * The discrimination weight(r) should be carefully chosen because it causes ignoring the first term when training(large r (e.g. r >= 1) makes the result bad) <p align="center"><img src="https://raw.githubusercontent.com/andabi/music-source-separation/master/materials/posen/disc_mse.png" height="30px"></p> <p align="center"><img src="https://raw.githubusercontent.com/andabi/music-source-separation/master/materials/posen/disc_kl.png" height="30px"></p> ## Experiments ### Settings * [MIR-1K dataset](https://sites.google.com/site/unvoicedsoundseparation/mir-1k) is used. * 1000 song clip with a sample rate of 16KHz, with duration from 4 to 13 secs. * extracted from 110 Karaoke songs performed by both male and female amateurs. * singing voice and background music in different channels. * Data augmentation * circularly shift the singing voice and mix them with the background music. * 1024 points STFT with 50% overlap (hop size=512 points) * L-BFGS optimizer rather than gradien
Excerpt of 9,172 characters
Read on GitHub
Dabi Ahn · South Korea
37
2
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:086a87d6815634f1, llm:Description: 'Deep neural networks for separating singing voice from music written in TensorFlow' (Python).
matched fp:086a87d6815634f1, llm:Description: 'Deep neural networks for separating singing voice from music written in TensorFlow' (Python).