alva/felsokning/docs/modules/communication-model.md

Module: Communication model (voice)

Canonical version. Swedish: communication-model.sv.md.

Core principle

The user speaks, the system writes.

The system uses voice-to-text for all spoken input. The technician should never have to type on a keyboard while work is in progress.

No voice agent: the system does not hold a running spoken conversation, does not read long answers aloud, and does not try to imitate a human conversation. The communication is speech in, text out.

Important design principle

All voice is treated as an input method, not as a separate interface.

All logic in the system is built on text. Voice-to-text is only a way of producing that text. That makes the solution easier to maintain, easier to search, easier to export, and easier to develop further with new models in the future.

Workflow

1. The technician presses the microphone: "I've measured between pin 14 and ground. I get 12.4 volts."
2. Voice-to-text transcribes the speech.
3. The transcribed text is sent to the model as an ordinary text request.
4. The answer always comes back in writing: Verified: supply voltage present at pin 14. Next step: check the ground connection at pin 7.

Why this choice?

• it works better in noisy workshops,
• it produces a permanent text log with no extra step,
• it makes the history easy to search,
• it reduces the risk of misunderstanding compared with a continuous spoken conversation,
• it suits cases where several technicians work on the same job.

Push-to-talk (PTT)

Voice input works on a push-to-talk basis. The app listens only while the user actively holds the microphone button, or after they have started an explicit recording. No background listening. No automatic activation.

Flow

1. The user holds down the microphone button (or presses a clear "Record" button, depending on the platform).
2. Recording starts immediately.
3. The app shows clearly that recording is in progress: a red indicator, a timer, a level meter, and the text "Recording".
4. The speech is transcribed in real time — the user watches the text appear and gets immediate feedback on whether the speech was understood correctly.
5. When the recording ends, the transcribed text is shown in an editable text field.
6. The user can accept, edit or re-record.
7. Only when the user confirms is the text sent onward and saved to the work log.

Editing before sending

The transcription is always editable. Common corrections: registration numbers, serial numbers, component designations, personal names, technical terms.

"Send" never happens automatically. The technician always gets a quick chance to correct the transcription before it becomes part of the permanent work log. That reduces the risk of incorrect registration numbers, component designations and measured values.

No hidden functionality

The user must always be able to see:

• when recording is in progress,
• when it has ended,
• what will be sent,
• what has actually been saved.

There must never be any doubt about when audio is being recorded or when information is being sent.

Automatic record keeping

Every transcribed sentence automatically becomes part of the work log:

08:14  "Measured voltage between pin 14 and ground. 12.4 volts."
08:14  System: Supply voltage verified.
08:15  "Relay doesn't click."
08:15  System: Check the control signal to the relay.

Everything is saved without the technician having to type a single line.

Hands-free working

The app is optimised for busy or dirty hands. During a normal case the user should be able to identify the object with the camera, photograph components, dictate observations, be shown the next step and carry on working — without typing manually. The interface has to work with gloves, dirty hands, strong sunlight, noise and vibration; the microphone button is large, easy to hit and gives clear visual feedback.