Voice quality
Whether the voice said the right things clearly can only be judged by listening. This page says what a listener would check, most safety-critical first. It measures none of it.
What a listener has to check, most important first
All six are Not measured. The order is by what a pilot acts on first, with the altimeter setting and the winds at the top.
1. Altimeter setting Not measured
It is the first number a pilot sets. Read wrongly, it changes what the altimeter shows, and the error may not be noticed until it matters.
What would measure it: A person listens to a real recording and compares the altimeter setting they hear with the one in the written briefing, digit by digit.
2. Winds: direction, speed and gusts Not measured
Wind decides runway choice and whether a flight is sensible. Direction, speed and gusts are three numbers that can each be misheard.
What would measure it: A person listens for each of the three and compares them with the written briefing, including how a gust is announced.
3. Ceilings and visibility Not measured
They decide whether a flight can be made under visual rules.
What would measure it: The same comparison of what is heard against what is written.
4. Hazards and restrictions: said, or left unsaid Not measured
When a hazard or a temporary flight restriction is present, or a check could not run, the caller has to be told.
What would measure it: A person checks that each hazard in the written briefing was spoken, and that a check that could not run was said to have not run.
5. Airport and frequency identifiers Not measured
A misheard identifier can send a briefing to the wrong airport.
What would measure it: A person checks how identifiers and frequencies sound and whether the caller had to repeat them.
6. Pace, pauses and interruptions Not measured
A rushed or halting delivery can hide the numbers that matter.
What would measure it: A person listens for pace, long silences and how the voice handles being interrupted.
What is checked today, and what is not
| What | State |
|---|---|
| The wording the service generates for speech | Exercised by automated tests in the service's repository. The tests read text. They do not listen to audio. |
| What a caller hears: the audio itself | Not measured Nothing on this dashboard checks it. |
| Call recordings | The real service archives them, and its admin page can play them. The Recorded demo calls tab plays a fixed snapshot of them, unedited. Nothing there checks how they sound. |
| Whether the voice said so when a lookup could not run | Not measured Needs a transcript or a listener. |
The first and third rows come from reading the repository and are not verified against the live service.
How a listening review would work
This is a plan, not a feature. No part of it exists on this page.
- Draw a small sample of real recordings each week, in a private, authorised review tool.
- For each call, a person records what they heard for the altimeter setting and the winds, then the rest of the list above.
- Any difference between what was heard and what was written becomes an item in the Improvement queue, with the recording as its evidence.
Related: Q-09, Voice quality is not measured.
Engineering details: where a recording would come from
The real call log has an audio_archived flag on a call record, and its admin page plays an archived recording. The synthetic figures do not read the call log. The recorded demo calls come from one fixed snapshot of it, taken beforehand, and this dashboard reads nothing from the log itself.
Nothing in this demonstration's events (fs.telemetry.request.v0) carries speech, text of speech or any measure of audio. The events are timing records for server requests.