DO's and DON'Ts

This topic describes some general dos and don'ts to improve the overall recognition of voice input.
 

Prevent Similar Answers

To easily improve recognition, it is recommended that you use answers for a question that differ audibly and do not overlap, meaning that one answer is a subset of another answer). Here are some examples: 

• [Active] and [Inactive] - Instead of using these, replace them with [Active] and [Disabled].
• [Left] and [Left Side] - Use [Overall Left] and [Left Side] instead.
• [Frequent] and [Frequently] - Use [Frequent] and [More Frequent].
• [First] and [First and others] - Use [First] and [Others].

Prevent Long Captions and Answers

The system will automatically use the full caption and answers configured, which means that when you use a long answer the system will only recognize an answer when the end user has provided the foll answer. For example:

• Lateral Occipitotemporal Gyrus
• Dimethylamidophenyldimethylpyrazolone
• Tetramethyldiaminobenzhydrylphosphinous acid
• Select the first three most applicable symptoms the patient has suffered from since the last visit

Add shorter, explicit speech commands to ease recognition and to increase the speed of answer.
 

Prevent Complex Captions and Answers

In some cases question captions or answers are expanded with the abbreviations, hints about how the answer should look, or alternative meanings or wording. This also makes the speech recognition more difficult as the end user must still say the entire caption or answer. For example:

• Previous registration date (year/month/day):
• Tumor size (cm):
• First, second + third (ABC)
• Vital signs are stable (VSS)
• DNC, D&C, or D and C

Again, use the explicit speech command configuration to improve recognition.
 

Custom Question Styling

The Assessment Framework can be customized and the part templates can be modified for a different kind of look and feel. Highlighting of the active/selected question and recording are controlled using a few style classes: groep_control, hgroep_control, flow_groep_control and horizontal_line_item. These classes will display the selection_border style class when a question is selected and the selection_border_recording style class when the recording is active. To define custom styling, keep in mind that these classes are necessary if highlighting needs to work out-of-the-box.
 

Custom Topbar

When customizing the topbar, remember to include the speech button if speech needs to be enabled/disabled using the topbar. The speech button is available in the speech_ui form include.

Prevent Keeping Recording Active

Currently, the most efficient recognition is obtained when using the press-and-hold configuration. After answering a question, briefly release the recording button to allow the system to fully process the results. If the recording remains active, a slight interruption is introduced in the recording while the server processes the answer. This means that the system may not recognize the next spoken information correctly because of the interruption. Wait until the answer is processed before answering a new question.
 

Logical Flow

Because speech input does not require that the end user constantly look at the screen, it is important to order questions logically. For example, group the most commonly answered questions together to prevent needless jumps over questions that do not need to be answered because they have a default value or because they are not mandatory.