About TargetOptimizer

TargetOptimizer 1

TargetOptimizer 1 is a free and open-source PC software written in C++ by Patrick Schmager to estimate pitch targets according to the Target Approximation Model by Yi Xu. Hence, it is similar to qTAtrainer, but differs in the following ways:

A 5th-order linear system is used as the filter to approximate the targets instead of a 3rd-order system.
The target parameters are jointly optimized instead of for one target after another.
The optimization includes a regularization term to penalize extreme target parameters, e.g., extreme positive or negative target slopes or time constants.
The estimated targets can be written to a gestural score file for VocalTractLab. This simplifies the copy-synthesis of natural utterances using the articulatory speech synthesizer.

The motivation for these differences are described in this paper.
To estimate the pitch targets for an utterance, the model needs as input the syllable boundaries in terms of a Praat TextGrid file, and the pitch contour in terms of a Praat PitchTier file. Based on these data, the pitch targets can be estimated and exported in terms of a CSV file, a gestural score for VocalTractLab, or as a Praat PitchTier file with the model f0 contour. The software can be executed as a command line tool (without a GUI, to support batch processing) or as an application with a GUI. To call the program with GUI, simply start it without any parameters. A screenshot of the GUI is shown below (for the German word "Ästhetik").

The vertical black lines are the syllable boundaries, the red lines are the estimated targets, and the green dotted line is the resulting pitch contour. The pitch samples of the original utterance are shown as blue dots.
To start the program as command line tool, call "TargetOptimizer.exe -h" to show the usage.

TargetOptimizer 2

The TargetOptimizer 2 is an extended version of the TargetOptimizer 1 and was written by Paul Krug, Simon Stone and Alexander Wilbrandt. It includes the optimization of the target boundaries and is not restricted to f0 trajectories. The additional optimization of the target boundaries can often improve the fit between the original and the modeled trajectories.

For full details on the various improvements over version 1, see:

Krug P, Stone S, Wilbrandt A, Birkholz P (2021). TargetOptimizer 2.0: Enhanced Estimation of Articulatory Targets. In: Studientexte zur Sprachkommunikation: Elektronische Sprachsignalverarbeitung 2021. Dresden: TUDPress. [pdf]