把你心情哼成歌

「把你心情哼成歌」（Sing Your Feelings, SYF）是一個結合文字辨識和音樂生成的系統，把一段有意義的句子、詩詞或文章轉換成一段好聽的音樂，並能體現出這段文字的意涵。

Files

- data/
    - raw/
        - *.mid
        - *.txt
    - csv/
        - *.csv
    - train.jsonl
    - test.jsonl
    - word-vectors.txt
- model/
    - {model name}.para
 
- crawler/
- gui/
    - index.html
    - ...

Requirement

Python3.5

• PyTorch should be installed manually from the website.

• Jseg3 should be installed by

  pip install https://github.com/amigcamel/Jseg/archive/jseg3.zip

Other packages are listed in requirement.txt, so:

pip install -r requirement.txt

midicsv

wget http://www.fourmilab.ch/webtools/midicsv/midicsv-1.1.tar.gz
tar zxvf midicsv-1.1.tar.gz
cd midicsv-1.1
make
make install INSTALL_DEST=path_to_install

Usage

Prepare Data

make

This will generate in the data/ folder:

• the word vectors word-vectors.txt

• the dataset train.jsonl and valid.jsonl

Train Model

python3 train.py

The trained parameters will be saved in model/test.para.

Server

python3 server.py

Modification

Train

The arguments are stored in config.py. One can find detailed descriptions about the arguments in it.

Server

• You can choose your own model, port and cuda usage in server.py.

  para = 'model/test.para' # path to your trained model
  model.use_cuda = False
  port = 80