# US 2021/0279457 A1 · Digital Handwriting Synthesis

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0279457 A1
- Document ID: US20210279457A1
- Status: Published application
- Kind: A1
- Application number: 16/813,144
- Filed: 2020-03-09
- Published: 2021-09-09
- Earliest claimed: 2020-03-09
- Assignees: Adobe Inc.
- Applicants: Adobe Inc.
- Inventors: Suyash
- Web page: https://patentgrep.com/patent/US20210279457A1

## CPC

- G06V 30/32: Digital ink
- G06F 3/04883: for inputting data by handwriting, e.g. gesture or text
- G06F 17/18: for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)}
- G06N 3/0475: Generative networks
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G06V 30/347: {Sampling; Contour coding; Stroke extraction}
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/08: Learning methods
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Digital handwriting synthesis techniques and systems are described that are configured to process text represented using text fields into one or more digital ink strokes to represent the text as handwritten data when rendered in a user interface of a computing device. Additionally, the digital handwriting synthesis techniques are configurable using normalization parameters to adjust an output from a machine learning model such that these techniques are extensible across a wide range of machine learning models and may be used to support a wide range of different digital ink styles. Further, the techniques described herein also support customization via “few shot conditioning” in which the digital ink styles may be further customized based on a user input and in this way support previously unseen digital ink styles.

## Claims

### Claim 1

In a digital medium handwriting synthesis environment, a method implemented by a computing device, the method comprising:
receiving, by the computing device, text entered via a first user input;
receiving, by the computing device, normalization parameters specified by a second user input, the normalization parameters corresponding to a particular digital ink style;
synthesizing, by the computing device, digital handwriting data based on the text by a machine-learning model, the synthesizing including generating stroke prediction data describing at least one digital ink stroke based on the text that is adjusted to correspond to the particular digital ink style based on the normalization parameters; and
outputting, by the computing device, the digital handwriting data as configured for rendering the text in a user interface by a display device.

### Claim 2 (depends on 1)

The method as described in claim 1, wherein the text includes a plurality of characters represented, respectively, using text fields and the generating of the stroke prediction data is performed concurrently for the plurality of characters in the text together by the machine-learning model.

### Claim 3 (depends on 1)

The method as described in claim 1, wherein the machine-learning model is configured as a transformer machine learning model.

### Claim 4 (depends on 1)

The method as described in claim 1, wherein:
the machine-learning model is implemented using decoder layers of a transformer machine learning model; and
the normalization parameters are incorporated by the machine-learning model using layers disposed between the decoder layers of the transformer machine learning model.

### Claim 5 (depends on 1)

The method as described in claim 1, wherein the second user input selects an option from a plurality of options of digital ink styles displayed in the user interface.

### Claim 6 (depends on 1)

The method as described in claim 1, further comprising updating the normalization parameters based on a third user input received via the user interface specifying a digital ink sample.

### Claim 7 (depends on 1)

The method as described in claim 1, wherein the generating the stroke prediction data includes:
generating a plurality of probability distribution functions as predicting a series of outcomes; and
sampling the plurality of probability distribution functions to generate the at least one digital ink stroke.

### Claim 8 (depends on 7)

The method as described in claim 7, wherein the series of outcomes describes movement in an X/Y direction to a target point.

### Claim 9 (depends on 1)

The method as described in claim 1, wherein the digital handwriting data as rendered appears as a handwritten version of the text.

### Claim 10 (depends on 1)

The method as described in claim 1, wherein the digital handwriting data as rendered appears as digital paint strokes.

### Claim 11 (depends on 1)

The method as described in claim 1, wherein the at least one digital ink stroke is specified in the digital handwriting data using a pen-down indication and a pen-up indication.

### Claim 12

In a digital medium handwriting synthesis environment, a system comprising:
an input module implemented by a computing device to receive at least one user input text specifying text and a selection of a digital ink style;
a style selection module implemented by the computing device to obtain normalization parameters corresponding to the digital ink style based on the selection; and
a machine-learning module implemented by the computing device to synthesize digital handwriting data, the machine learning module including a machine-learning model configured to generate stroke prediction data describing at least one digital ink stroke based on the text and adjust the stroke prediction data based on the normalization parameters.

### Claim 13 (depends on 12)

The system as described in claim 12, wherein the text includes a plurality of characters represented, respectively, using a plurality of text fields and the generating of the stroke prediction data is performed concurrently for the plurality of characters in the text together by the machine-learning model configured as a transformer machine learning model.

### Claim 14 (depends on 12)

The system as described in claim 12, wherein the machine-learning model is a transformer machine learning model.

### Claim 15 (depends on 12)

The system as described in claim 12, wherein:
the machine-learning model is implemented using decoder layers of a transformer machine learning model; and
the normalization parameters are incorporated by the machine-learning model using layers disposed between the decoder layers of the transformer machine learning model.

### Claim 16 (depends on 14)

The system as described in claim 14, further comprising an updating module implemented by the computing device to update the normalization parameters based on a third user input received via the user interface specifying a digital ink sample.

### Claim 17 (depends on 14)

The system as described in claim 14, wherein machine-learning module is configured to generate the stroke prediction data by:
generating a plurality of probability distribution functions as predicting a series of outcomes; and
sampling the plurality of probability distribution functions to generate the at least one digital ink stroke.

### Claim 18

In a digital medium handwriting synthesis environment, a system comprising:
means for receiving at least one user input text specifying text having a plurality of characters and a selection of a digital ink style;
means for synthesizing digital handwriting data for the plurality of characters in the text concurrently by a machine learning model, the machine-learning model configured to generate stroke prediction data describing at least one digital ink stroke based on the text and adjust the stroke prediction data based on the normalization parameters.

### Claim 19 (depends on 18)

The system as described in claim 18, wherein the machine-learning model is implemented using a transformer machine learning model.

### Claim 20 (depends on 18)

The system as described in claim 18, wherein:
the machine-learning model is implemented using decoder layers of a transformer machine learning model; and
the normalization parameters are incorporated by the machine-learning model using layers disposed between the decoder layers of the transformer machine learning model.
