Suppose a person cannot speak [/and ;\ wants to send a voiced message to someone.
The current granularity of selecting a voice |may not be sufficient.
Two approaches could exist:
A. For people that can see and also for blind people (just use voice to activate the special features (using special voice keywords), for each syllable can be underscored by a note symbol chosen (just bracket the syllable and place a note as a subscript, each letter can also be given a phoneme type (thrilled r or rolled r, easy h, harsh h, or cough h, (or other experimental sounds)), machine sounds for letters as subscripts as well (nice, medium, and worse - sounds, perhaps only pronouncable by a machine but linguistically nice-sounding). Also select a vowel length via a subscript.
B. Let the computer sound the word in a variety of possible ways, and choose from the list. Good for people that can hear, and also for blind people who can say "ok" or NEXT, at each step, but not good for people who can'tspeak or could have a speech disorder.
C. This method is not exclusive of A and B, but could be used in conjunction:
Place the letters along the time axis, spaced on the time axis as you like. You can also fix them if you cannot move arms, mouth, but only move your eyes. With your eyes. Blink, stare at keyboard, stare at letter, blink to select, user interface switches to graph, stare at time axis (a special medium-gray circle moves slowly left or right, as you stir left or right, blink to fix letter along time axis (you will need a large screen interface here, perhaps an Android TV monitor). Quickly blink five times to proceed to y-axis. Again, you will place your dark-gray circles along t-axis, then quintuple-blink (more then 5 blinks will act as 5), move each vertically to control tone (in the same way as on x-axis). Right-eye blink: go back. Left-eye blink: go forward. Long-blink: quickly jump forward along action and subaction tree. Double long-blink: SOUND IT. THREE CONSECUTIVE long-blinks: SEND.