Workbench User Manual for Transcribers

Introduction
AppTek’s Workbench is a cloud-based media production platform that allows you to transcribe, edit, QA
(review and proofread), and search audio and video files. The Workbench integrates with AppTek’s deep
learning Automatic Speech Recognition (ASR) and neural Machine Translation (MT) technologies to
provide automated closed captions (transcripts) and subtitles (translations) in multiple languages. The
virtualized framework allows for a distributed workforce to simultaneously review the output, perform
post-editing and implement quality control to ensure accuracy, flexibility and affordability.

Based on different transcription use cases, transcribers and editors also have the option transcribe from
scratch (without ASR) and with or without automatic segmentation. Transcribers will also have the option
to label speakers.

Workbench Access
1. Go to: http://apps.apptek.com:20000
2. Log in using your username and password provided.
3. Upon logging in, you will see the Programs List that displays all the files uploaded to your Site.
This will show the Name (file name of your audio), Description (domain/topic), Dates added and
modified, Language, percent Transcribed, percent Translated and percent QA’d.
Workbench Navigation
1. You may filter the list of Programs to quickly find your file by searching for the file name or
specifying a date range of when the file was uploaded. To activate the search, click on

2. The Action icons on the left will allow you to edit, review or export each file. To transcribe or edit
a file, you will click on the blue pencil icon.

3. Editor View

Waveform: At the top of the screen, you’ll find the audio Waveform. As the audio plays, the
Current Cursor Point will move through the waveform. This shows you exactly where you are in
the audio.
Media Options and Features:

1. Loop Icon: Click this Loop icon so that it’s yellow. When this is enabled, the audio will play
only the current segment. The audio will stop when it gets to the end of the segment and
not continue to play to the next segments. When this icon is not selected (gray), then the
audio will play continuously throughout the file.
2. Split: Click this button when you want to split up the segments to create a new segment.
3. Show Markers: By clicking this button, the system will display red arrows at the bottom
corners of each segment. You can click and drag these red corners to adjust the segment
boundaries to shorten or elongate the segments.
4. Play: this button plays/pauses the audio
5. Playback Speed: You may choose to change the playback speed of the audio to slow the
audio down (helpful when a speaker talks fast) or speed the audio up.
6. Zoom bar: Dragging this to the right will increase the size of the waveform, so that you can
more precisely segment the audio. Dragging this to the left will decrease the size of the
wave form so you can see an overview of more segments at once.
7. Timestamp Range: This lists the start and end points of the current selected segment.
8. Segment Duration: This displays the current segment duration in seconds.
A. My Role: When assigned a file to transcribe, you will select the “Transcriber” role from the
dropdown menu. Certain users will have the permissions to select QA. When assigned a file
to QA, you will select the “QA” role from the dropdown menu. Any corrections a QA’er
makes during this mode will be saved and analyzed to measure and score the work of the
original transcriber for word error rate accuracy.
B. Shortcuts: Click this button to display keyboard shortcuts.
C. Speaker Data: This button will allow you to modify speaker name, globally specific to the
selected speaker. Also, you can enter speaker attributes, such as gender and dialect.
D. Approve Record: This button will allow you to mark a file as Complete.
E. Validate: Click this button to ensure proper transcription rules are being following. It will
list any errors in tags, numbers, restricted characters, or filler words. Transcribers should
validate their transcripts often to ensure accuracy of transcription.
Segmenting and Transcribing
Shortcut Keys
[CTRL] [S] Manually Save
[TAB] Play and pause the audio
[SHIFT] [TAB] Rewind audio 2 seconds
[CTRL] [ENTER] Split segment (insert segment breakpoint)
[CTRL] [SPACE] Non-speech tags (see #4 below)

1. Prepare your transcription file: Click on the Loop Icon, “Show Markers”, and moving the zoom
bar to the right to increase the waveform size.

2. Play the Audio: Start playing the audio at the beginning of the audio file by pressing Tab on
your keyboard. You can point your cursor to any point in the waveform to start playback. Or,
you can point your cursor to any point in the text segment to play the audio at the beginning of
that particular segment.
3. Transcribe: Listen to the audio and start typing what you hear, word for word. You may press
Tab again to pause the audio. The system will automatically save your work every 10 seconds.
If your segment has orange bars at the beginning and end of the segment, this means that your
work is NOT saved! Please make sure your internet is connected and that you don’t leave the
page and lose your work. Tutorial video to recognize any unsaved work:
https://youtu.be/52PW6hV2Pz8
4. Tagging: Please follow the Transcription Guidelines closely for specific tagging conventions. To
speed up your work in inserting tags, you may use the following shortcuts:
a. Non-speech tags: start typing the first 3 or 4 letters of the tag (without the squared
bracket) then press Ctrl Space to auto complete the tag. Tutorial video:
https://youtu.be/jYiKCKJYThQ
b. Initialism tag: highlight the acronym or initialism, right-click, then click on the <initial>
</initial> to insert before and after the highlighted word. Tutorial video:
https://youtu.be/JP9Kws7o2zc
c. Code-switching tag: highlight the text in the foreign language, right-click, then click on
the <lang:foreign> </lang:foreign> tag to insert these tags before and after the
highlighted text. You may edit the word “foreign” to the actual foreign language of the
audio. Tutorial video is same as above: https://youtu.be/JP9Kws7o2zc
5. Segmentation: Transcription requires accurate segmentation and each segment should be less
than 15 seconds. Segments need to be based on speaker turns. Every time a different speaker
talks, you must segment the audio at that point. If a speaker talks for a long time (over 15
seconds), then you will segment the audio at a sentence level or pauses in speech.
a. Adjusting Segment Lengths: Fix the segment lengths by clicking on the red triangle at
the bottom of the segments and dragging it to the left or right.
 b. Split Segments: Point your cursor on the waveform at the precise spot you want to split
the segment. Then, point your cursor to the text within the segment where you want to
split the text. Click on the Split button, or do Ctrl Enter on your keyboard.
c. Create Segments: The only way to create a segment is to SPLIT a segment into two
segments. Follow the step above to do this.
d. Merge Segments: You can merge and combine two segments together in two ways.
From the waveform, you can drag the red corner arrows all the way into the next
segment (overlapping the two segments) and click OK to merge. For the transcript, you
can click on a text segment, hold the Shift key down, and then click on the other
segments that you want to merge it with. This will highlight the segments in blue. You
will Right-click, and select Merge Selected.

6. Speaker Labeling:
a. You must label the speaker for each segment. Speaker labels should be consistent
throughout the entire file. You will type the speaker label at the text box at the start of
each segment. Press Enter after typing the speaker name. If you don’t know the
speaker’s name, then you can type a generic name (Reporter1, Reporter2, Male1,
Male2, Speaker1, Speaker2, Doctor, Waiter, Boy, etc.).
b. If there is overlapping speech for over 3 seconds, segment this overlapping speech on
its own segment and put the speaker label “multiple”.
c. Each speaker will be saved for each file, so that when they talk again, you can select that
speaker label from the dropdown menu. Please note: Speaker labeling, once selected or
entered, you must click outside of the speaker label box to begin transcription for that
segment.
d. *Speaker Data*: You must identify speaker attributes/metadata for each speaker. This
includes identifying the gender (male, female, or unknown) and dialect (native or non-
native) for each speaker. If the speaker is speaking in a dialect other than the main
target language, you will need to type the dialect spoken.

7. Validation: To ensure transcription is following our specific rules, click on Validate to check if it
passes our transcription validator. The validator checks to see if a transcriber forgets to spell
out numbers, or forgets to remove restricted special characters, or misspells non-speech tags. If
there are style errors, then an orange box will appear with an error count.
Click on this orange box to highlight the segments with errors. You can see in the example
below, the number 6 appears in the transcript, which isn’t allowed.

Note: Validation might not work for all languages at this time.