Вы находитесь на странице: 1из 2

Instructions for Masterscript Proofreading Task

Overall
The aim of this task is to verify that the reading scripts we have
created for speech data collection purposes are correct, and to
highlight any problematic items that may lead to issues in recording
so that these can be addressed.
Background
All of the unique phrases that people will be asked to read aloud in
our recordings have been grouped together into a masterscript,
which is on the first tab of this file.
These scripts were generated by taking various inputs and
combining them in order to cover the requirements of the client. The
prompts were created by taking prompt patterns (column E), such
as "album <AlbumName>" and inserting relevant values into the
slots (dynamic or changeable content - column G), such as "Bad" or
"Lucie". This would lead to the following types of prompts: "album
Bad" and "album Lucie".
Although all the individual inputs to the scripts were checked, we
often find that once all the possible prompt permutations have been
created, certain errors become apparent that were not obvious
originally. For this reason, we would like to have all the prompts
checked to ensure that they will not present issues at the recording
stage.
These scripted prompts are designed to mimic commands that you
would say to your in-car speech recognition system. They cover
functionalities such as navigating to particular places, calling
contacts, searching and playing various media, etc. As these
commands are designed to reflect what a person may say to a
machine, some items may sound a bit unnatural, such as:
"Call Ola home"
or

"Play song Strawberry Fields Forever"


Items such as this should not be considered incorrect. However, if a
prompt sounds very bizarre and you think it might lead to confusion
during recordings where speakers are asked to read it, please
highlight it and provide a comment explaining why.

Instructions for "proofreading"


The file is very large, and it doesn't make sense to proofread the
whole thing in detail, as often it is simply a pattern that needs to be
corrected. The following approach is recommended to complete this
task efficiently:
1. Read fairly quickly through all the prompts in column
B (you do not need to proofread any other columns), and highlight
in an obvious colour any items that are problematic (2-2.5 hrs).
"Problematic" prompts are those that:
a. Contain misspelled words
b. Do not make sense, or are ungrammatical (with the
exception of the "unnatural" types described above)
i. this can include incorrect grammatical case
c. Contain content that is not valid, e.g.:
i. a country name spelled in English or another
language
ii. an address that contains a foreign city or street
iii. a phone number does not follow a valid phone
number pattern (phone numbers will include
international/emergency numbers)
iv. a name that is very uncommon or difficult to
say
2. Go back to the highlighted items and provide a correction or
comment explaining what the issue is (1-1.5 hrs).
a. You can copy and paste the same correction or
comments to multiple cells if you are correcting a repeated pattern.
i. Please note that in these cases, you can also
provide a correction for the relevant "prompt pattern" in column E.
NB: For song names, artist names, audiobook names, etc, foreign
titles are permitted.

Thanks, and good luck with the task!

Вам также может понравиться