The travel assistant of Nordlicht Reisen GmbH answers in the portal and on the phone, according to R-02. The phone variant is, according to E-06, a voice-based prompt agent through Voice Live. Learn lists it as a preview: The lesson builds no production commitment on it. In addition: How does a model understand a recording directly (R-04), and how does spoken Danish become readable (R-01)?
Prior knowledge. Prompt agents, versions and conversations: lesson 2.3. Recognition and speech output with the Speech SDK: lesson 4.3. Text translation: lesson 4.2. Foreign content as data and hard rules in code: "AI track, lesson T6.2" and "AI track, lesson T6.4".
New in this lesson. The phone agent is called reiseassistent-telefon and is an agent of its own beside reiseassistent: Learn says that a text agent cannot be switched over into a voice agent. It starts with German and English; Danish on the phone is open (see below). Recordings are made up (R-12).
Paper, account, costs. Paper: decision and exercises; an Azure account is needed only for the optional run. Cost drivers according to Learn: model, speech recognition and output, Application Insights, telephony connection (Communication Services or Twilio), connected tools; units on the pricing page for voice agents.
The requirement
All numbers are assumptions of the task.
| No. | Requirement | Type | Measure | Weight | Source |
|---|---|---|---|---|---|
| 4.4-01 | Callers talk to the assistant without typing. | F | conversation in German and English | Should (preview) | R-02, E-06 |
| 4.4-02 | Refund, legal matters and blocked account go to a person. | R | 0 exceptions in the test set | Must | R-02 |
| 4.4-03 | Recordings and texts are stored only when necessary. | R | storing off by default | Must | R-12 |
| 4.4-04 | A voice message can be read as a summary. | F | file up to 20 MB | Should | R-04 |
| 4.4-05 | A Danish message becomes readable in English for staff. | F | 1 target language, text | Should | R-01 |