Community Wishlist Survey 2017/Archive/Voice edit

Voice edit

N Out of scope for Community Tech

Problem:

Reading Wikipedia by (smart) mobile phone is convenient for many use cases. However editing Wikipedia on mobile phone is hard, partly because of the keyboard on (smart) mobile phone. Voice input in mobile phone would be a lot faster, but some specific functionality - such as wiki link - is not support natively.

Who would benefit:

Most editors, who may find voice input is easier in a more mobile / ubiquitous computing world, in major languages with Speech-to-Text (STT) supports advanced enough (Vietnamese, my native language, is one of such languages)

Proposed solution:

Support voice input for mobile - integrating with default voice input of the mobile OS (Android - Google service by default, iOS - Apple service by default) or can be setup by user to use their preferred STT service .

By "support voice input", I mean, for example, Wikipedia application can allow certain "voice syntax" for wiki link. The syntax, beside default definition, can be setup by user to their preference.

More comments:

Phabricator tickets:

Proposer: Tttrung (talk) 08:20, 10 November 2017 (UTC)[reply]

Discussion[edit]

Consider regional languages too-- Shagil Kannur (talk) 05:10, 11 November 2017 (UTC)[reply]

An interesting idea. Worth to look deeper into it. If the devs think thay can make it work, why not? --Vachovec1 (talk) 18:34, 12 November 2017 (UTC)[reply]

The way voice editing experiences like this have historically worked is that the person records what they want written, and then someone else transcribes it and makes any corrections necessary. Nowadays, voice recognition is generally good enough that the first person can use text-to-speech for automated transcription, but it will always require human checking due to the error rate. The only areas where human checking is not required are areas where small errors are not critical, and only short bits of speech are recorded to allow the user to easily correct themselves, such as Alexa, Siri, Duolingo, and similar. A fully text-to-speech driven environment, where the user also uses text-to-speech to correct previous errors, is possible, but tends to be so cumbersome to use that people only use them if they have no choice, such as blind users. Developing software solutions for this that work in a performant manner on mobile devices in the web browser is extremely hard, which is why the operating systems of phones typically include this in an integrated fashion. General text-to-speech solutions already exist in most modern mobile phones, and even in older or lower-spec phones used in emerging communities. Making a system that allows the user to use wikitext on top of that adds extra complexity, even ignoring the user customisation aspect. In short, this proposal has obvious value, but I think it's such a large undertaking that I don't see it happening within the scope of this wishlist. --Dan Garry, Wikimedia Foundation (talk) 12:35, 17 November 2017 (UTC)[reply]

At WMSE we plan to have a project that will result in a tool for collecting speech data. This would be a first step in creating an open source ASR-solution for MediaWiki. It will be a follow-up to our current project Wikispeech, that allows the user to have an article read out loud. Keep an eye on the page for updates! /Sebastian Berlin (WMSE) (talk) 13:52, 20 November 2017 (UTC)[reply]

Yes, as Deskana (WMF) said above, this is too big for the Community Tech team. I'll have to archive. Thanks for participating in the survey. -- DannyH (WMF) (talk) 19:33, 20 November 2017 (UTC)[reply]