fishandchipsing / brds

A not-so-intelligent intelligent thing

brds

A not-so-intelligent intelligent thing

Install

Build a virtual environment so we do not step on toes

virtualenv venv
source venv/bin/activate

When you are finished be sure to deactivate the virualenv

deactivate

Install requirements

python -m pip install -r requirements.txt

Install fluidsynth

brew install fluidsynth

Install pyfluidsynth https://github.com/nwhitehead/pyfluidsynth

Download the MIR-QBSH-corpus and move it inside the /Dataset folder

wget http://mirlab.org/dataSet/public/MIR-QBSH-corpus.rar

Add a recon/ folder or else you will get a directory not found error

Install timidity, needed for auto conversion to wav

brew install timidity

Download the soundfont to the folder https://packages.debian.org/sid/fluid-soundfont-gm

Create the folder Dataset/soundfonts and place fluid_r3_gm2.sf2 in it or any other Soundfont you like

Create the folder Dataset/recordings to old the saved files.

Run standalone code

Record for five seconds humming and then translate to midi and play back (it will)

python birds.py

On line 257 of birds.py you can switch to custom Sound fonts subprocess.call(['fluidsynth', 'Dataset/soundfonts/fluid_r3_gm2.sf2', midi_f, '--no-shell'])

#subprocess.call(['fluidsynth', 'Dataset/soundfonts/Birdsongs_Arizona.sf2', midi_f, '--no-shell'])

Use existing .wav humming recording from the MIR-QBSH-corpus

python birds.py --fname 'Dataset/MIR-QBSH-corpus/waveFile/year2003/person00001/00014.wav'

Notebook Version

jupyter notebook

Open the query-by-humming notebook, and run interactively from there.

Run

python decomp.py

brds's Issues

provide endpoint wrapper API

rather than have a jupyter notebook, provide a simple python wrapper to handle the basic functionality of pitch detection.

Characterizing a Dataset

Lets start with a unified dataset of bird sounds that we can characterize nicely; some attainable number in the hundreds of short (less than 30s) clips of clear sounds.