Choosing a speech-to-text approach

TypeWell is an ideal tool for providing communication access and notes to a great majority of students and others. There are other communication access systems, however, including sign language interpreting, oral transliteration, and CART.

If you have decided that a speech-to-text approach may be right for a particular student, you'll then choose between different types of speech-to-text solutions, including:

• TypeWell or other meaning-for-meaning systems;
• Communication Access Realtime Translation (CART, also called realtime stenography);
• Voicewriting (based on automatic speech recognition), which has two subtypes:
  • Professional voicewriters, where the operator has taken courses and/or received a degree in speech-to-text voicewriting, practiced for months or years, and has possibly been certified through the National Verbatim Reporters Association (NVRA) at a performance and accuracy level.
  • Casual voicewriters, such as a teacher in a classroom whose speech goes directly to the computer, or an employee in the department who uses off-the-shelf speech recognition software on a day-to-day basis.

Of these three, TypeWell requires the least training (as little as 40 hours spread over 8-12 weeks), is operable by the largest pool of potential transcribers, and produces the highest quality output where meaning-for-meaning transcripts are desired.

CART (also known as stenotype) requires the most training (2-5 years), is operable only by a relatively small pool of careerist writers, and produces the highest quality output where verbatim transcripts are desired.

Professional voicewriting requires a significant amount of training (1-2 years), is operable only by a relatively small pool of careerist writers, and produces near-verbatim output similar to CART.

Casual voicewriting is not a viable service option because the quality is not high enough for communication access. It is listed here for comparison's sake. Casual voicewriting requires little training (about 60 hours to reach the "moderately experienced desktop dictation" level), and is operable by a medium-sized subset of people with the right voice and temperament. Unfortunately although the amount of training is similar to that of TypeWell, the high error rate makes this potential service type unacceptable for communication access.

It's important to check the training level of the service provider for any speech-to-text service, but is absolutely crucial for verbatim services like CART and voicewriting. Because demand is so high and the training so difficult, sites might be tempted to hire people still in training, or training dropouts. It takes years of intense training to become good at stenotyping. Someone only partially trained is likely not able to provide adequate service. You wouldn't hire an interpreter with no credentials, and it's wise not to do so with speech-to-text providers, either.