real-time pitch correction t-pain on a...
TRANSCRIPT
![Page 1: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/1.jpg)
6.111 Final Design ProjectTrevor Walker and Ishwarya Ananthabhotla
T-Pain on a Chip:Real-time Pitch Correction
![Page 2: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/2.jpg)
Autotuning•Pitch correction-–Fixing intonation of audio signal–Without changing other features–Post-processed•Everywhere!–Contemporary vocal artists–Auto-generation of harmonies–Interesting effects in instrumentals
![Page 3: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/3.jpg)
Motivation
•Signal processing in frequency shifting?•Investigate challenges: sampling, resolution, quality of sound, “naturalness”•Limitations of a hardware?•Realtime?
![Page 4: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/4.jpg)
Project Goals and Expectations
•Realtime vocal autotuning system•Corrects pitches so that they match user entered scale – i.e. “C Major”•Features–Degree of correction–Rate of correction–Range of upper male to female vocal frequencies–Preserve reasonable output quality
![Page 5: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/5.jpg)
Block diagram
![Page 6: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/6.jpg)
FFT
AudioCodec
•Window audio signal•Sampling rates meet minimum frequency specs
•Fixed point, low-speed FFT•18 bit precision, up to 27Mhz, 4% consumption of FPGA•Read and Write across SRAM-like interface
![Page 7: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/7.jpg)
Phase vocoder
FFT bins are too wide to determine frequency accurately
Exploit phase information in FFT to drastically increase resolution ht
tp://sethares.engr.w
isc.edu/vocoders/phasevocoder.html
![Page 8: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/8.jpg)
Phase vocoder, continued
Phase is related to frequency: f ∝ dθ/dt
Resolution of phase comes from precision of arctan, not number of FFT bins
By observing the change in phase between successive FFTs, we can figure out the true frequency
![Page 9: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/9.jpg)
LUTs
Note LUT converts frequency (440Hz) into human language (A4)
Shift LUT takes fundamental, produces:● Closest note in specified scale● Ratio fdesired/factual
![Page 10: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/10.jpg)
Shifter
Shifts FFT based on LUT result (multiplication)
Rate of change of pitch shift limited by correction speed
Degree of correction controls magnitude of shift
![Page 11: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/11.jpg)
IFFT
AudioCodec
•Entire spectrum shifted based on scaling of fundamental frequency•Non-normalized, manually scale
•Convert back to audio waveform•For output through speakers
![Page 12: Real-time Pitch Correction T-Pain on a Chipweb.mit.edu/6.111/www/f2014/projects/ishwarya_Project_Design... · Trevor Walker and Ishwarya Ananthabhotla T-Pain on a Chip: Real-time](https://reader034.vdocuments.mx/reader034/viewer/2022050323/5f7c3725c1aa257f04626fbd/html5/thumbnails/12.jpg)
Timeline
Week 1 (Nov 12) Week 2 (Nov 17) Week 3 (Nov 24) Week 4 (Dec 1) Week 5 (Dec 8)
FFT/ IFFT Ishwarya
Vocoder
LUTs Trevor
Shifter
Interfacing Both
Final Report/ Adjustments