Singing Synthesis from MIDI File
The midi2voice project offers an intriguing venture into the realm of singing voice synthesis. With its roots in the advanced tech of the Nagoya Institute of Technology's sinsy.jp website, midi2voice utilizes a Hidden Markov Model (HMM)-based Singing Voice Synthesis System to breathe life into MIDI files, transforming them into synthesized singing voices.
Requirements
To make the most of midi2voice, users need to meet a couple of basic requirements:
- MuseScore: This software is used to convert MIDI files into MusicXML format.
- Python 3: The project is built with and runs on Python 3, a modern, high-level programming language.
Installation
Getting started with midi2voice is straightforward. By running a simple pip install command, users can integrate this tool into their systems:
pip install git+git://github.com/mathigatti/midi2voice.git
Usage
Utilizing midi2voice is user-friendly—it can be run from the command line with a series of flexible parameters to tailor the output to specific needs. The command includes options to input lyrics, specify the MIDI path, choose language and gender for the voice, adjust tempo, and much more.
Here's a basic structure to run midi2voice:
python -m midi2voice -l SHALLOW.txt -m SHALLOW.mid
For users who need guidance, a help command is available:
python3 -m midi2voice -h
Usage Example
To illustrate its capabilities, users can check out sample MIDI and text files, which are thoughtfully provided here. These examples help users to try and see how midi2voice turns a simple MIDI and text file into a synthesized singing performance.
Try it on Colab
For those without a local Python setup, midi2voice is accessible online via Google Colab. This option makes the project easily available for experimentation without any installation requirements, providing a quick and hassle-free platform for testing.
Available Voices
The tool offers a variety of voice options across different languages—Japanese, English, and Mandarin. Each language supports several female voices and a few male voices (except for Mandarin which currently lacks male options). Users can specify their choice using the VOICEINDEX parameter, ensuring a customized singing synthesis to suit different projects or preferences.
Additional Information
The midi2voice project is a creation of Mathias Gatti. Recognizing that many might find this tool beneficial for academic and practical applications, Mathias Gatti also encourages citations for scientific publications using the provided DOI link. Contributions to further support this open-source initiative can be made through donations, helping continue the development and maintenance of such innovative projects.
By offering tools like midi2voice, Mathias Gatti exemplifies the power of collaborative, open-source efforts in advancing technological capabilities and widening access to sophisticated music synthesis technology.