Speech to Text Devices

This pay-as-you-go subscription software has a range of features which work on any cell phone. Users can create and send 15-second text messages using their voice. It can also be used to update social networking statuses, and even create short “Twitter”-style captions, which are then synced with your personal Internet accounts via wireless or a GPS (Global Positioning System).

Vlingo Plus is speech-recognition software for BlackBerry users. Rather than a pay-as-you-go system, this requires a basic on-off purchase. The program allows users to tell their BlackBerry phones to text, email, call and update statuses with short pieces of text via a side key. This key is easy to locate by touch even when you cannot see the device properly, such as in the dark or when you are jogging.

BigHand is a speech-to-text program designed and implemented by United Kingdom-based company BigHand. Users record their voice using what is basically a dictaphone inside a mobile communications device (such as a smart phone or BlackBerry). The voice recording is then sent to a speech recognition workflow and the sounds (words) are transcribed into text, optionally with proofing, which adds accuracy and readability. It is useful for detailed text, such as law documentation and note-taking. Any mistakes can be corrected by the user, as described by BigHand.

Dragon is a versatile dictation program, made by Nuance, which allows users in a variety of professions to create and edit data in documentation and email without having to type – they simply make the changes using voice commands. Also, full pieces of work can be developed using dictation rather than typing, simply by turning the program on and speaking into the microphone as it records. Dragon offers speech-driven clinical documentation and communication for the healthcare industry. Dragon is also being used to improve customer interactions for a variety of communications companies and call centers.

To get the most from any speech recognition device or software program, the program must be trained to recognize a minimum number of real voices. This is because the algorithms within the software work by learning the nuances of the person’s voice and accent over time. If the device is used by one person or a small group of people, it can have an accuracy of 85 percent, but this falls rapidly for larger numbers of users. Recognition software also works by learning from its mistakes, so it's always wise to perform manual corrections following a translation.