Tutorial for UCLA Voice Synthesizer

A tool for manipulating the source of sound developed by Jody Kreiman, Bruce R. Gerratt, and Norma Antoñanzas-Barroso.
Citation: Kreiman, J., Gerratt, B. R., & Antonanzas-Barroso, N. (2006). Analysis and synthesis of pathological voice quality. Unpublished users’ manual.

Program: here (Google might regard it as a malware and prohibit you from downloading it. Try other browsers.)
Manual: here
Example sound: here
Example output: here

Open a WAVE file (Note: The sampling rate should be 10000 Hz. You can resample your sound file in Edit → Downsample.)
Select a clip of the file that contains at least 8 cycles.
Click “FFT” button (the default setting is usually fine). Output: Spectrum in the bottom left.
Click “LPC” button (the default setting is usually fine). Output: The filter envelop shows above the spectrum.
Mark a period on the waveform. Left click to mark the left boundary; Right click to mark the right boundary.
Click “F0” button. Output: The F0 value shows up on the top right corner of the waveform.
Click “IF” button to run the inverse filter. Output: Top right: Glottal waveform (in black) and Derivatives of the glottal waveform (in magenta). Top bottom: Source spectrum.
To smooth the derivative: Modify the formant frequency and bandwidth values. The smoother the derivative, the better.
Save: File → Custom Save → Save for Synthesizer → Windows. The working directory should be the folder where you saved the .exe files. Folders of “SYNTHESIS” and “WORK” are automatically created. You can find the output from the Invf.exe in a folder with the same name of the original waveform.

File → Open case → Open the .lv file generated from the Invf.
Variables → Initialize (Default settings are usually fine. The LPC order need to be the same as in Invf.exe. The default is 14.)
Source → LF → Find LF features automatically.
Source → LF → Fit an LF model.
Click “Display audio waveform” button. Mark the beginning of a cycle.
Move cursor to the start of a cycle and Click “Pitch” button. Output: Each cycle is marked with red line.
Click “FM” button. Output: Model the F0 contour.
Click “AM” button. Output: Model the Amplitude contour.
Click “N” button. Output: Model the noise.
Click “S” button. Output: Synthesize the voice.

Mark the spectral tilt: Mark the left harmonic: Shift + left click Mark the right harmonic: Shift + right click
To change the value by dragging:
- Double left-click the red dot
- Move the dot up or down
- Double right-click the red dot
To change the value by assigning values of spectral tilt:
- Type a value in “Source Spectral Slope”
- If want to change the value at the left of the tilt: Ctrl + left click
- If want to change the value at the right of the tilt: Ctrl + right click
To make multiple harmonics smoothed to the tilt:
- Point to the middle dot of the spectral tilt
- Press and hold the right mouse button
- Press and release the left mouse button while holding the right mouse button

Tremor D: Tremor Deviation; Value of the synthetic tremor frequency deviation from F0 selected with the slider
Tremor F: Value of the synthetic tremor cut off frequency selected with the slider
Jitter D: Value of the synthetic jitter deviation selected with the slider
Shimmer D: Value of the synthetic shimmer deviation selected with the slider
Noise (Noise to signal ratio): Value of the synthetic noise selected with the slider

File → Save all
- Either save the file to the same folder as the .lv file, or create a new folder to save everything.
The synthesized output: .syn (1 second maximum)