Francis Tyers

We should keep track of tokenisation so sentences can be merged

5

For example here:  The original text is: ``` common_voice_th_23657260.mp3 สองอันเท่าไหร่ ``` e.g. without spaces.

Sometimes the audio is at very different volumes

2

Not sure if this is anything we can do something about, but the audio coming out of Common Voice is sometimes at very different volumes...

It would be good to have feedback buttons for sentences where people can (optionally) say if they understood the sentence or not. This would help us calibrate which sentences should...

Another one from @brianrocca: * "Odd one out". So click on the word that was different. Just use 3 different tokens to make it a perception activity. Make it harder...

Suggested by @brianrocca: - http://www.manythings.org/ac/ Basically: - Memory game for the audio. One idea would be to give people longer segments and get them to remember and pick out as...

A task for identifying N words in the audio

6

The task: * Play an audio * Get presented with e.g. 6 words (or Chinese characters) * 3 are in the audio, 3 are not (distractors) * You have to...

It would be great to give people the opportunity to learn a writing system too. Here is my idea: * Have the audio * Take the transcript, split into characters...

Put the instructions behind a click

1

Maybe something like [`fa-question-circle`](https://fontawesome.com/icons/question-circle?style=solid) from FontAwesome.

Choose or type based on image

1

It would be cool to also allow people to learn by typing in images or by choosing between N words for a given image. To avoid the huge data collection...