Possibilities on improving the quality of recorded clips in CV

Common Voice
#1

Hi!

Here some thoughts and proposals of improving the quality of recorded clips
(on the server side, “pushing the user in the right direction”).

  • Every recorded clip, who makes it in the corpus gets a point.

  • For Troll clips (2 second empty mouse click clips, and so on…) negativ point(s) are given.

  • The guidelines/rules for recording clips could be saved in a textfile for every speech section
    and must also be read/recorded by every user.
    (I did not read the Faq/Guidlines is no excuse anymore)

  • This is also possible for validating clips (guidelines for validating clips are saved in a textfile for every speech section.
    Disadvatage would be: anonymous users who are deleting their cookies are in a loop by reading/validating always the same clips. Maybe someone else has a solution for this (contributing clips only with account?)

  • Before the user can upload his clip(s), he MUST hear his own recordings. Microphone issues and so on are recognized by the contributor himself.

  • To get more voices from children and teens:
    Placing posters in primary and secondary schools?

1 Like
#2

Hey robovoice,

I like the idea of the point system, but I am not certain if Mozilla is willing to implement an entirely big new feature like this. Some sort of feedback abou the quality surely would be great. In the past, you could see your percentage of accepted sentences in the profile. They deleted this feature for performance reasons, but I think it was very useful.

This one looks easy to implement and would have a huge impact. Some people tent to make errors when they donate for too long, like switching words or they click too late on record and cut the beginning of the sentence. Forcing them to hear everything at least once would increase the quality a lot.

1 Like
#3
  • A way of Feedback for the contributer in the form of a Summary/Overview of recorded and uploaded/rejected/skipped/successfully validated sentences.
    This could be indicatet with coloures in the summary:
    yellow - uploaded but not verified yet
    green - uploaded and positive validated
    red - uploaded and rejected sentence, try
    again
    blue- correct sentence which is included in the corpus
    black - skipped sentence/reported sentence

Usefull for non native speakers/contributers with account for speach building/improving, who are doing the same errors over and over again.

For contributers with smartphones and account:
Splitting up the sentences in 500/1000 blocks?

#4

Please make sure you have read this:

1 Like
#5

I also opened some requests/bugs on github about my proposals on improving this situation.

2 Likes