synthviz

synthviz is a library for creating visualizations of piano-playing from MIDI files. The videos look like this:

Right now, synthviz just provides a Python API. Command-line API is hopefully coming soon!

Requirements

system requirements

You'll need to install a couple of tools that make rendering this video possible:

ffmpeg (creates video from audio and image video frames) - on Ubuntu, sudo apt-get install ffmpeg
timidity (synthesizes piano audio from MIDI) - on Ubuntu, sudo apt-get install timidity

python package requirements

Install this package via pypi:

pip install synthviz

Usage

You can use synthviz through the Python API:

from synthviz import create_video

create_video('river.midi') # provide str path of MIDI file

Options

The create_video function provides a lot of of options:

def create_video(input_midi: str, 
        video_filename = "output.mp4",
        image_width = 1280,
        image_height = 720,
        black_key_height = 2/3,
        falling_note_color = [75, 105, 177], # default: darker blue
        pressed_key_color = [197, 208, 231], # default: lighter blue
        vertical_speed = 1/4,
        fps = 20
    )

input_midi (str): path to MIDI file
video_filename (str): path to output video, synthviz will write the video here
image_width (int): width of output video in px
image_height (int): height of output video in px
black_key_height (float): height of black keys as a percentage of piano height
falling_note_color (Tuple[int]): color of falling keys in video, list of three RGB integers
pressed_key_color (Tuple[int]): color of pressed-down keys in video, list of three RGB integers
vertical_speed (float): the speed of the falling keys, fraction measured as main-image-heights per second
fps (int): frames-per-second of output video

Creating video from raw audio

With the help of the piano_transcription_inference library, you can make a cool video directly from raw audio!

First, install that library via pip install piano_transcription_inference. Then run the following code:

import librosa
import os
import pathlib

from piano_transcription_inference import PianoTranscription, sample_rate, load_audio
from synthviz import create_video

audio_input = 'my_audio.mp3'
midi_intermediate_filename = 'transcription.mid'
video_filename = 'output.mp4'

transcriptor = PianoTranscription(device='cuda', checkpoint_path='./model.pth')
audio, _ = librosa.core.load(str(audio_input), sr=sample_rate)
transcribed_dict = transcriptor.transcribe(audio, midi_intermediate_filename)
create_video(input_midi=midi_intermediate_filename, video_filename=video_filename)

Credits

The synthviz library was originall adapted from this blog post, "Making Synthesia-style videos in Ubuntu", written by David Barry. Thanks David!

jxmorris12 / synthviz

readme