26/08/2021, 19:09 Loft 2.

0 User Guide  |  Google Developers

warning Confidential material:

Loft 2.0 User Guide bookmark_border

First things first

You are required to park the question as soon as it appears in your workbench, before you
start to transcribe

Click on the P button to park the question.

When you click on the Park button, the dialog shown below will appear. Select the option
shown in the screenshot and click the Park
button outlined in red.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 1/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Layout

The page is partitioned into four main areas:

The task bar - displays icons that can be clicked to view Shortcuts or Guidelines as
well as to skip and submit tasks.

The canvas - displays the waveform and the transcribed segments.

The workspace - where transcription work is done

The left panel - displays controls to select speakers and labels, which will affect the
view of the canvas and workspace.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 2/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Canvas

The audio controls for the audio waveform are on the top section of the page.

Directly above the audio waveform, are the play controls, the audio counter in the format
mm:ss:ms and the length of the audio in hh:mm:ss. The blue line on the waveform indicates
where the audio is playing/paused currently. The audio controls include:

Play/Pause button - Plays and Pauses the audio. The space bar on the keyboard can
be used as a shortcut.

Rewind button - Will bring the blue line back to the starting point of the audio and also
pause the audio when clicked.

Largest Gap button - Will jump the blue line forward or back to whatever gap between
segments is the largest in the entire audio timeline.

Audio counter - indicates the progress of the audio in minutes (m), seconds (s), and
milliseconds (ms).

Playback Speed dropdown menu - allows the user to modify the speed they listen to
the audio.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 3/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Cursor

The blue cursor indicates the exact point of the audio that you are currently listening to and
corresponds to the counter.

As audio is played, the cursor, waveform of the audio, and the canvas will scroll to the right.
The cursor cannot be dragged to a different part of the waveform, but you can click and
drag the horizontal scrollbar directly below the waveform to position the cursor on a
different part of the waveform of the audio.

When the Play button is clicked, the audio will begin playing from the point in the audio that
the blue cursor has been positioned on the waveform of the audio.

Workspace

The workspace is where the transcription is completed. Each transcription box shows the
following fields for a segment that need to be set as you transcribe:

The start time

The end time

The speaker

The text of the transcription for the segment

The transcription box also contains:

The Save button

The Cancel button

The Save button saves the entered transcription. You will find that the speaker name is on
the canvas and when you hover over the speaker name, the transcription will be visible.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 4/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

The Cancel button will undo any unsaved changes made in the transcription box. However,
if previous changes were saved, the button will be greyed out.

If you modify the segment and click outside the transcription box, a pop-up will instruct you
to either save or discard the changes.

Save Changes

Left Panel

The Left Panel displays controls to select speakers and labels, which will affect the view of
the canvas and workspace, based on what is selected in the respective sections.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 5/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Add Transcription/Label Segment

Add Segment

The + in the rainbow circle is the Add Segment button. This creates a segment that is 1
second long. A shaded segment box appears on the waveform and a transcription box
appears for the segment in the workspace area. When you click the +, a segment named
“No speaker” appears in the workspace and the canvas under the corresponding section of
the audio. This transcription box is immediately in focus with a cursor in the text box.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 6/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Add Content to the Segment

1. Click on the segment in the canvas, this will take you to the transcription box for that
segment in the workspace.

2. Click on the “No Speaker” in the transcription box. This will show a dropdown menu
that has an “Add Speaker” option and an option to select any previously created
speakers.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 7/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

3. Click the “Add Speaker” option, which will prompt you to enter a “Speaker Name” and
select a color for that speaker. The color will be used to denote the speaker on the
canvas and workspace every time the speaker is selected.

Edit Segment

https://developers.google.com/speech-data-ops/user_guides/loft2.0 8/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

You can either set the transcription start and end time by dragging the segment in the
canvas.

You can reopen the transcription box for a saved segment by clicking on the segment on
the canvas or in the workspace.

Delete Segment

Click the X on the upper-right corner of the segment in the workspace to delete a saved
segment.

Split Segment

The Split button allows you to split a selected segment.

1. Click inside the segment where the split should occur on the waveform. This moves
the cursor to that location and opens the transcription box in the workspace below.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 9/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Note: You can split the transcription text by placing the cursor in the desired spot to
divide the transcription. The first half would go into the first segment and the second
half would go into the next segment.

2. In the transcription box, click the Split button for the corresponding segment.

The original transcription (or the first half if you clicked within the transcription text) will
move to the first segment box generated.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 10/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Speakers

Once a speaker has been added in Loft, their name will appear in a dropdown menu in the
transcription box when you create a new segment. Selecting an existing speaker from the
left-hand menu will render their segment to the canvas under that speaker's name.

To create a new speaker, you can either click the + symbol in the left-hand menu under
Speaker or create a segment as defined earlier.

As segments for different speakers are added, they will be differentiated by the color and
indentation of segments so that all segments for a particular speaker will be represented
on the canvas in the same color under that speaker's name.

Labels

https://developers.google.com/speech-data-ops/user_guides/loft2.0 11/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Some projects in Loft will require labels. Check the project-specific guidelines to know
which labels to use.

To add a label:

1. Set the cursor to the appropriate location for the label on the waveform.

2. Click the + icon.

3. In the new segment created in the workspace, modify the label field and then click
Save.

The label will render under the appropriate speaker label if the label is:

Foreign Speech

PII

Address

ID: email, phone

Singing

Unintelligible

The labels below will render outside of speakers' segments since they are not attributed to
a speaker label.

Applause

DTMF: Dual-Tone Multi-Frequency

Laughter

Music

https://developers.google.com/speech-data-ops/user_guides/loft2.0 12/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Noise

Ringtone

Unknown

To remove an label, click on the already created label and click "remove" on the label box.

Other Functionality in Loft 2.0

Prefill

Prefilled audio will have some transcription automatically generated into the workspace.
These prefilled transcriptions will have the following note: “Content was pre-filled and needs
approval”.

All prefill segments will need to be edited and saved before you can submit the task.

Shortcuts

Shortcuts button is available on the upper right-hand corner and clicking on this button will
display the keyboard shortcuts available.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 13/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Useful functions

https://developers.google.com/speech-data-ops/user_guides/loft2.0 14/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Listen to all of the audio from one speaker

Change the filters in the left panel to hide undesired segments. You use the next/previous
segment buttons to skip between segments by this speaker.

Replay the audio for a specific speaker segment

You can replay the speaker segment by hitting the replay button.

"Parking" Questions

Workers are required to park the question as soon as it appears in their workbench, before
starting to transcribe.
A worker is allowed to park a maximum of 1 question.
This ensures
that this question will stay assigned to the same worker for the entire time that
trasncription is in progress. The new Parking functionality allows workers to stop and
resume work on a given question at their convenience. While the question is parked, it will
not be assigned to another worker. A worker who has started work on a question will have
access to it until it is completed. Questions can be parked for ten (10) days.

Once a question is parked, the icons show that it is parked. The rectangle on the left shows
the number of days left for it to remain assigned to the worker.

The grayed out parking button in the screenshot below shows that the worker has parked 1
question and hence is unable to park any more questions.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 15/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Review

When a worker completes and submits a question, the question is added to a review
question type. The submitted question appears in the review question type with the original
transcription as prefill. The reviewer can accept the prefill (i.e. the transcription by the
original worker) or make edits and then submit.

FAQS

Review FAQs

1. As a Worker, how can I determine which Question Types are Review Question Types?
*. As shown in the picture below, the “Question Type’ will be tagged with _review.

Parking FAQs

1. How do I use the Parking feature?

Workers are automatically assigned questions on the workbench (referred to as

Worker Parks Question 1 by selecting the “Park” button.

Worker is assigned a new question (Question 2).

https://developers.google.com/speech-data-ops/user_guides/loft2.0 16/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Workers should click on “Parking Lot” immediately, select Question 1 that was
just parked and, when prompted, select “Don’t park the workbench question and
open the parked question”

Outcome: Question 1 stays assigned to the worker. Question 2 may not stay
assigned to the worker.

2. What happens if a Worker does not see the “Park” button? The worker either:

Already has 1 question parked, so the “Park” button will be gray and instead they
should load a question from the Parking Lot.
OR

There are no more questions left in the pool, so they cannot see the “Park”
button.

3. Can a Manager unassign Parked questions?

Yes. If necessary, the manager can unassign Parked questions from a Worker’s
Parking lot.

Scenario: A Worker is unable to work on a question assigned to them.

Manager logs into CrowdCompute Manager URL.

Manager selects the pool and sees the Parked questions.

Manager select the question Parked by that Worker and selects “Un-
assign”.

The question is now returned to the pool. Note: Any and all previous work
WILL be lost.

4. The Reporting Tool says there are questions remaining in the pool, but the Worker
sees that “No more questions are available.”

a. The MANAGER can navigate to the Reporting Tool and record the total
number of questions in the pool (For example, 10).

b. The Manager can check the number of completed questions in the pool (For
example, 8).

c. The Manager can check how many questions are Parked (For example, 1).

d. Verify that the workers are working on a Parked question. If the Worker
follows the instructions in (1), they will be working on a Parked question.

e. Number of unassigned questions.

Add the numbers in 4b and 4c.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 17/18
26/08/2021, 19:09 Loft 2.0 User Guide  |  Google Developers

Subtract the number obtained in the bullet above from 4a.

(4a - (4b + 4c))

(10 - (8 + 1)) = 1 unassigned question

All rights reserved. Java is a registered trademark of Oracle and/or its affiliates.

Last updated 2020-10-14 UTC.

https://developers.google.com/speech-data-ops/user_guides/loft2.0 18/18