Handwriting synthesis results

We showcase handwriting samples synthesized by the proposed method. The style inputs and the generated outputs are rasterized as images for illustration.

Handwriting, nonparallel text

We showcase nonparallel-text online-handwriting synthesis. The style example is shown first in each row.

content input	Meetings will also remain private.
style input
generated output
content input	Please raise your hand.
style input
generated output
content input	Tell me what to do
style input
generated output
content input	I want to sing a song
style input
generated output
content input	We have the easy part
style input
generated output
content input	I would like to ask you out
style input
generated output
content input	All he does was sleeping
style input
generated output
content input	Sometimes the sky becomes blue
style input
generated output
content input	I hope he never comes back
style input
generated output
content input	still water runs deep
style input
generated output

Handwriting, parallel text

We showcase parallel-text online-handwriting synthesis. The style example is shown first at each row.

content input	This is a very common
style input
generated output
content input	Moreover, it must be for
style input
generated output
content input	They were very unfair to
style input
generated output
content input	These days are long gone
style input
generated output
content input	Who will attend?
style input
generated output
content input	That all hope is never
style input
generated output
content input	The judge was really nice
style input
generated output
content input	It should be equal
style input
generated output
content input	I have the first six
style input
generated output
content input	What was the matter for
style input
generated output

Handwriting, random styles from prior distribution

We showcase the capability of the proposed method to sample random styles from the learned prior distribution. For Graves [8], we remove priming and sample from the output distribution. As can be seen, both methods can synthesize handwriting with random styles, but the proposed method learns a prior distribution of style, whereas in Graves[8] the randomness of style is embeded in the output distribution.

Content input 1:

This is a big, big win.

generated output

Content input 2:

It's gonna be alright!

generated output

Handwriting, interpolating between two styles

We showcase the capability of the proposed method to sample random styles from the learned prior distribution. For Graves [8], we remove priming and sample from the output distribution. As can be seen, both methods can synthesize handwriting with random styles, but the proposed method learns a prior distribution of style, whereas in Graves[8] the randomness of style is embeded in the output distribution.

Input text 1:

We are back to square one.

style 1
interp coeff = 0
interp coeff = 0.25
interp coeff = 0.5
interp coeff = 0.75
interp coeff = 1
style 2

Input text 2:

Please don't go away

style 1
interp coeff = 0
interp coeff = 0.25
interp coeff = 0.5
interp coeff = 0.75
interp coeff = 1
style 2

*Due to privacy reasons, the style references used in this video are synthetic. They are similar to unseen real style examples in our dataset and are synthesized using a generative model with a different architecture. The generations shown here are very similar when real samples are used as style input. Note that all the evaluations reported in the paper are done using real unseen style examples.