TL;DR
This paper introduces "Access For All" (AFA), a free, web-based Danish text-to-speech service designed to make online information accessible to individuals with reading difficulties. The service has proven viable and popular, demonstrating a clear need for such tools and the sufficiency of current internet infrastructure to support them.
What Problem Does This Paper Address?
Approximately 30% of the Danish adult population experiences significant reading difficulties. With the increasing amount of information available online, this poses a democratic challenge, potentially excluding a substantial portion of the population from essential knowledge. The AFA service aims to bridge this gap by providing an easy-to-use, free tool that reads aloud any marked text on web pages, ensuring "equal access" to information for everyone.
Methods
The AFA service utilizes a client-server architecture. A small, downloadable Windows application (the "remote control") allows users to mark text on any webpage. This marked text is then sent to a web server hosting a Danish text-to-speech synthesizer. The server analyzes the text and converts it into synthetic speech, which is then returned to the user as an MP3 audio file for playback. The design emphasizes simplicity and transparency for the end-user, with the remote control featuring only basic "replay" and "stop" buttons, though later versions added pitch and speed adjustments. The server infrastructure is based on HP Netserver hardware running Windows 2000 Server.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- The "Equal Access" Imperative: The paper strongly highlights how the internet, while a powerful tool for information dissemination, can exacerbate existing inequalities if not made accessible to all. AFA directly addresses this by democratizing access to online content.
- Viability of Free, Web-Based TTS: The success and sustained usage of AFA over 28 months demonstrate that a free, online text-to-speech service is not only needed but also technically feasible with current internet bandwidth and server technology.
- User-Centric Design is Crucial: The emphasis on a simple, intuitive interface for the "remote control" is key to AFA's adoption. Users with reading difficulties need tools that don't add cognitive load.
- Response Time Trade-offs: The paper details the careful balance between speech synthesis time, MP3 compression, and network transfer time. While synthesis is the bottleneck, MP3 compression is necessary for efficient transfer, even if it adds processing overhead.
- Diverse User Needs: The analysis of text themes reveals that AFA is used for a wide range of purposes, mirroring the diverse interests of the general population, from public authorities and news to entertainment and private communication.
Key Results
- High Initial Adoption: Within the first week of launch, the AFA website saw over 7,000 unique visitors, with approximately 3,500 downloading the remote control. This number grew to over 10,000 downloads in the first month.
- Sustained Usage: By the end of 2004, the service was handling around 5,000 daily text-to-speech conversions, with an average of 184 unique remote controls connecting daily.
- Average Response Time: The average response time from submitting text to audio playback is approximately 7 seconds.
- Text Synthesis Time: The speech synthesizer takes an average of 5 seconds to convert 404 characters into a WAV sound output.
- Diverse Usage Patterns: The service is primarily used on weekdays during typical working hours, suggesting use in educational or professional settings.
Limitations and Future Work
- Dependence on Public Funding: A significant challenge is the service's reliance on public funding, with future work aiming to achieve greater independence.
- API for Commercial Integration: One idea being pursued is the development of an API (application programming interface) that can be sold to homepage owners, enabling them to integrate AFA's accessibility features directly into their sites.
- Multilingual Support: The paper identifies the need to support other languages, particularly English, given its importance in Denmark and interest from international institutions. This would require minor modifications to tag the language of the text.
- User Characterization: Due to the policy of not requiring user registration, it is difficult to precisely characterize the user base, though feedback from emails and telephone calls provides insights.
- End-User Perception Study: A systematic evaluation of end-users' perception of response times is still to be undertaken.
Summary updated