Nowadays, we are living a new accessibility paradigm where removing barriers is not the only way because of universal accessibility have became one of the main goals to be achieved. According to this principle, it is herein presented a system which eliminates hard of hearing students’ communication barriers in educational environment, present the information in an accessible format according to the Spanish regulation and creates digital content automatically for all students. It has been implemented a whole system which transcribes the teacher’s speech and converts the spoken lesson into a new digital resource, with the help of a voice recognizer. This transcription is available in real time for students with hearing needs in form of captions, shown in the main screen of the classroom (if desired) or sent wireless and available as closed-captions in accessible clients implemented in a laptop, personal computer or personal digital agenda. This information and the teacher's speech are stored in XML and WAV files for a latter use as support for the learning process via internet. The system allows the students with speaking problems to type in a keyboard their questions or comments that are sent back via radio to the teacher's computer where they will be converted to synthetic voice by means of a text-to-speech program.
Some parameters of this text-to-speech transformation, like speed or volume, can be controlled by the teacher's interface, to make the class accessible for students with reading problems.
The recognition process is very sensitive to training time and highly depends on the microphone used, so a personal implication of the teachers is needed. Recognition error rate achieves values below 5% with a training process of two hours. Once the desired error rate is achieved, and specific vocabulary added to teacher’s speech files, this process does not need to be repeated never again.
The system has been tested in a deaf integration school in Madrid with a great acceptation of both teachers and hearing impaired students. Other environments are taken into account whenever a training process can be reached, such as guides, conferences or theatre real-time captioning.
One limitation of this implementation is the voice recognizer, since it is a specific application based in the commercial Nuance’s Dragon Naturally Speaking program, so a license is needed. In order to reduce error rates, this program transcribes whole sentences, so a delay is introduced between the teacher’s speech and the moment when captions are shown. Depending of the length of the sentence, this delay can reach 10 seconds, but this is reduced with some speaking guidelines such as not to build long sentences, not to doubt, etc. If these norms are taken into account, the system does not trouble significantly the rhythm of the class. Another possible limitation is the need of modern computers, because of the speech recognition, and one per class. Finally, questions of hearing students does not pass through the system, so it is not registered, nor transmitted to hard of hearing students, unless the teacher repeats the question.