[DRAFT] Task Setting in Common Voice

This page is unofficial. Please refer to the forum and official website for accurate information.

• ref. Mozilla Voice Community Playbook

• The origin of the dataset is to collect texts.
• The text and voice corpus are the material of the dataset.
• Contributors to the text corpus are users of the Sentence Collector.
• Contributors to the voice corpus are users of the web app.
• Engineers who train the system are users of the dataset.
• The final end user is everyone (or the system's target population).

How to contribute

• Helping the Community――Help people start and continue to contribute to the project
  • Answer questions in the forum and chat
    • Forum: Common Voice - Mozilla Discourse
    • Chat: Common Voice - Matrix
  • Participate in the open discussion
  • Share feedback and ideas
  • Keep a diary of my activities
  • Share organized information
• Development support――Develop, fix and maintain sites and tools
  • Common Voice · GitHub: Sentence Collector, Sentence Extractor, Dataset, etc.
  • GitHub - mozilla/common-voice: Web app
  • Report a bug
  • Suggestions for features

Task setting before dataset generation

1. Add Language
   • Make a topic to request in the forum
     • ref. Readme: How to see my language on Common Voice
2. Translate Web App
   • How to do: Common Voice - Pontoon

Task Setting in Collecting Text

1. Find
   1. What? (Know what to collect)
      • Public domain (PD)
        • ref. Public domain - Wikipedia
      • CC0
        • ref. CC0 - Creative Commons
        • ref. CC0 FAQ - Creative Commons
   2. Where? How?
      • ref. How-to on Common Voice Sentence Collector
      • ref. Ideas for finding public domain text
      • ref. Text Corpus Link Collection
   3. If found: [Important] Check the rights status of the text (ja)
   4. How to add
      • Add by myself?
        • Collect
          • Manually?
          • Automatic?
      • Just sharing information?: Text Corpus Link Collection
2. Add
   • Check the writing style
     • ref. How-to on Common Voice Sentence Collector
     • ref. 文章の編集について（使用文字の確認）
     • Editing?
   • Add
     • Questions about sentences: 文章収集・CollectorツールのQ&A（自家製）
     • Review by myself?

   Memo

   • What rules do you think your language needs to have?
     • Japanese: Sentence collector for Japanese language (日本語の文章について)
3. Review
   • Review Criterion: Discussion of new guidelines for uploaded sentence validation
   • If we find a sentence that is not in the public domain: Sentence collector copyright issues

   Memo

   • The review will carry over to the previous and subsequent pages
   • What to do with a sentence we can' t decide on?: We can ignore it

   Question

   • What happens to the rejection sentence?

• What is the process for making changes to the collection?
• Personal Suggestion
  • Post #18 on Sentence Collector Open Discussions - Input needed - Common Voice - Mozilla Discourse

Task Setting in Collecting Voice

1. Recording
   1. Prepare an audio input device (microphone, etc.)
   2. Read the sentence
      • Questions to read: Common VoiceのQ&A（自家製）
   3. Check to see if the voice is recorded
   4. Submit clips

   Memo

   • What do we care about when we record?: Common Voiceの録音と検証のやり方
   • What to do with a sentence we can't or don't want to read?: Skip button
   • What to do with an inappropriate sentence?: Report button
   • We can submit less than 5 clips
   • We can delete clips in the middle of a recording to end it

   Question

   • What do deal with a microphone that doesn't work (can't input voice)?
2. Validate
   1. Don't look at the sentence, listen to the voice
   2. Next, look at the sentence and make sure it matches the voice
   • Validation Criterion: Discussion of new guidelines for recording validation
   • Questions on listening: Common VoiceのQ&A（自家製）
   • Common Voiceの録音と検証のやり方

   Memo

   • What to do with a clip we can't decide on?: Skip button
   • What to do with inappropriate voices and sentences?: Report button

• What is the process for making changes to the clip?
• Personal Suggestion:
  • How to record effectively?
    • Automatic volume detection: automatically fixes or presents the problem to the user
  • How to validate efficiently?
    • Automatic detection of un-validated clips
      • Case 1: No voice is recorded