This article is available in: English Svenska
Photo by Manfred Werner/Tsui, CC BY-SA 3.0.

Photo by Manfred Werner/Tsui, CC BY-SA 3.0.

Wikispeech, a so called speech synthesis, aims to make all the information available on Wikipedia accessible for anyone that faces reading difficulties, whether that is due to vision impairment, dyslexia, they never had an opportunity to learn literacy, or any other number of reasons.

Everyone should be able to take advantage of the information on Wikipedia. Only offering written text is really limiting for millions of people and this project we hope will be part of the solution to realize the grand vision of providing free knowledge for all people of the world.

The development of a text-to-speech MediaWiki extension was carefully investigated and planed during a pilot study in Autumn 2015. Development requirements and existing solutions that could be integrated were identified at this stage.

Just like everything else that the Wikimedia movement creates, the entire software will be freely licensed alongside the speech data and documentation. As such, people will be able to use Wikispeech on other wikis run by other organizations. Moreover, different parts of the synthesizer can also easily be reused by other text-to-speech projects, so that even more people with limited reading abilities will have an easier life.

The hope is that Wikispeech will be used in many different services, and that the communities improving the software and lexicon will be very diverse.

WikiSpeech logo. Photo by ElioQoshi, public domain/CC0.

Wikispeech logo. Photo by ElioQoshi, public domain/CC0.

Many of those most in need of information do not have easy access to a personal computer, have limited knowledge of how computers and smartphones work, and/or lack access to the high quality commercial tools that are available. To reach these people the aim is a MediaWiki extension. Even though an extension is much more difficult to develop than an independent external tool we hope that it will be easier to find for Wikipedia’s users and there will be no need to download any software as it is an Internet-based solution. Another benefit is that a MediaWiki extensions should be simpler to include in other extensions and tools developed by other teams, such as reading aloud of subtitles in videos shown on Wikipedia, in Kiwix (the offline Wikipedia), and elsewhere.

At the end of the project, our goal is to have Wikispeech available in Swedish and English, along with a working prototype in Arabic, to show that it works in a right-to-left language. Long-term, we hope that Wikispeech will be available in all languages.

It will be possible through crowdsourcing to improve Wikispeech. You may, for example, help develop the pronunciation dictionary used by the program to make it sound more correct (the pronunciation of each word is described by the International Phonetic Alphabet). In this way, the speech synthesizer is constantly improved and annoying errors can be corrected by the user. The project team is not aware of any similar solutions and believe that this might actually be a world’s first. Wikispeech will also use a very flexible structure, where different parts of the synthesizer (dictionaries, voice, etc.) easily can be connected. The goal is that a newly developed solution that has been missing simply can be plugged into the framework.

We want the value of Wikipedia to increase among the people who have the least resources and are most vulnerable. A speech synthesis has a very great value in many of the world’s poorer countries where the education level is low. For commercial entities, these are not markets that are a priority, and here Wikispeech can help to fill the gaps.

The Swedish Post and Telecom Agency is financing the work that Wikimedia Sweden (the project owner) and the project partners Södermalms Talteknologiservice AB (STTS) and the Royal Institute of Technology (KTH) are doing in the project. In total, the project has a budget of around US$300,000. Furthermore, a number of other organizations and research teams have offered to support the project in other ways, including reporting in more than sixty news articles worldwide.

To make things as transparent as possible and make it easier for volunteers to help out, show their support, or raise a note of caution, all of the planned development have been added to Phabricator. We invite you to get involved, and please endorse the project!

John Andersson, Wikimedia Sverige