AUTO-TUNE
Introduction
Auto-Tune, whose name well explains its functionalities, is a combination of multiple music processing techniques. It includes, but is not limited to, pitch correction, timeline correction, and de-noising. Auto-Tune is usually used to tune pitches to where they should be, often matching a standard 12-tone music scale. In addition, people can use it the other way around - tuning pitches away from the right position, creating a new music genre.
The focus of this project though, is pitch correction and de-noising. Timeline correction is trickier because it involves word detection. We'll leave it for now.
To do de-noising, we apply soft thresholding and block thresholding techniques. To do pitch correction, we apply a number of different algorithms with varying degrees of success. We attempt pitch correction with a ground truth (a desired pitch at each time) and without a ground truth (fitting pitches to the nearest note on a 12-tone scale).
Our approach uses the following DSP tools:
The focus of this project though, is pitch correction and de-noising. Timeline correction is trickier because it involves word detection. We'll leave it for now.
To do de-noising, we apply soft thresholding and block thresholding techniques. To do pitch correction, we apply a number of different algorithms with varying degrees of success. We attempt pitch correction with a ground truth (a desired pitch at each time) and without a ground truth (fitting pitches to the nearest note on a 12-tone scale).
Our approach uses the following DSP tools:
- Short-Time Fourier Transform and Wavelet Transform
- Soft, Hard, and Block Thresholding for De-Noising
- Phase Vocoding and Threshold Pitch Modification without Ground Truth
- Autocorrelation for frequency detection
- Column-based, Segment-based, and pitch-based shifting and PSOLA with Ground Truth
Work Flow
Step1: Collect noisy to-be-tuned data.
Step2: Perform Denoising on the noisy data.
Step3: Perform different methods of autotuning on the denoised data.
Step4: Get results, compare and analyze.
Step2: Perform Denoising on the noisy data.
Step3: Perform different methods of autotuning on the denoised data.
Step4: Get results, compare and analyze.
Team Members: Yufan Yue | Mingshuo Shao | Yuhan Chen | Eric Winsor
{funkyyue, mingshuo, chenyh, rcwnsr}@umich.edu
{funkyyue, mingshuo, chenyh, rcwnsr}@umich.edu