A reading voice that is not a punishment.

Accessibility is the use case a meter distorts most quietly. The alt text gets terser, the description gets skipped, the confirmation stops being read back — each one a small saving, and together the difference between a product someone can use and one they cannot.

24 kHz

default output — full rate, not the telephony rate

Where the meter edits accessibility

Every accessible affordance is extra characters: the longer image description, the read-back of a confirmation number, the heading structure spoken aloud, the error explained rather than announced. On a meter each of those has a price, and the pressure is always toward saying less.

On a seat the extra sentence is free, which removes the argument entirely. The decision returns to what a person needs to hear.

The whole text, every time

Film

The longer description, the spoken read-back, the error explained rather than announced. On a seat none of those is a saving anyone is tempted to make.

What matters technically

Spell tags read codes, references and confirmation numbers character by character, which is the difference between a usable read-back and a guess.
A low temperature holds a steady, predictable delivery — the right setting for a voice someone listens to all day.
Break tags let a structure be heard: a real beat between a heading and its content.
24 kHz output by default, so the audio is full-rate rather than the compressed telephony band.

Notes — an engineer's checklist

01Is this a screen reader?

No — it is the synthesis layer a screen reader or an application can call. Gandr turns text into audio over an API; the navigation model, the semantics and the focus handling stay in your product or in the assistive technology the person already uses.

02What sample rate does the audio come back at?

24 kHz mono PCM by default, which is the full-rate audio rather than the 8 kHz telephony band. Set output_sample_rate if your playback path wants something else.