# dl_signal

**Repository Path**: ngc13009/dl_signal

## Basic Information

- **Project Name**: dl_signal
- **Description**: No description available
- **Primary Language**: Python
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-01-08
- **Last Updated**: 2024-01-08

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Complex Transformer
A deep learning model which incorporates the transformer model as a backbone and develop complex attention and encoder/decoder network operating on complex-valued input. This model achieves state-of-the-art results on music transcription tasks. <br />
**Title:**      Complex Transformer: A Framework for Modeling Complex-Valued Sequence <br />
**Arxiv:**      https://arxiv.org/abs/1910.10202 <br />

## Requirement
python: `python=3.6`<br />
cuda  : `cuda>=9.0`<br />
packages: `pip install numpy scipy sklearn intervaltree resampy torch`<br />
This code base relies on GPUs and cuda.
## Transformer model
<p align="center">
  <img src="https://github.com/muqiaoyang/dl_signal/blob/master/img/transformer.png" alt="Complex Transformer"/>
</p>

## File parsing
`cd music/`<br />
`wget  https://homes.cs.washington.edu/~thickstn/media/musicnet.npz`<br />
`python3 -u resample.py musicnet.npz musicnet_11khz.npz 44100 11000`<br />
`rm musicnet.npz`<br />
`python3 -u parse_file.py`<br />
`rm musicnet_11khz.npz`<br />
`cd ..`<br />
(This process is quite long)

## Instruction
Preprocess the MusicNet dataset as stated in the paper: <br />
`python parse_file.py`<br />

Sample command line for automatic music transcription: <br />
`python -u transformer/train.py`<br />

Concatenated transformer for automatic music transcription: <br />
`python -u transformer/train_concat.py`<br />

For MusicNet generation tasks: <br />
`python -u transformer/train_gen.py`<br />

Concatenated transformer for MusicNet generation: <br />
`python -u transformer/train_gen_concat.py`<br />

For IQ classification task: <br />
`python -u transformer/train_iq.py`<br />

Concatenated transformer for IQ classification task: <br />
`python -u transformer/train_iq_concat.py`<br />

For IQ generation tasks: <br />
`python -u transformer/train_gen_iq.py`<br />

Concatenated transformer for IQ generation: <br />
`python -u transformer/train_gen_iq_concat.py`<br />

LSTM for MusicNet generation: <br />
`python -u lstm_music_gen.py`<br />

LSTM for IQ generation: <br />
`python -u lstm_iq_gen.py`<br />

## Path Configuration
Example: `python -u transformer_train.py --path PATH`
## Parameter Tuning
All the parameters you can tune are in the argparser section of train*.py or lstm*.py file.