CEPD · Module 4 — Scenario, dialogue, and avatar direction
What research says about synthetic voices and on-screen agents, how to make Arabic AI narration sound right, and the consent and disclosure rules a production director should enforce.
About 22 min
Practice in the lesson player
Sorting, matching, quizzes, and the module gate stay behind a free account so scored answers are not public. The teaching notes below are the same cited curriculum.
01
A pilot group says the Arabic narration "sounds like a robot". The voice is a good synthetic voice. The problems are elsewhere: it reads "SAR 5,000" as "S-A-R five thousand", stresses the wrong syllable in the client's brand name, uses a news-reader tone for a friendly coaching line, and pronounces a Gulf place name the Egyptian way. None of these is the voice engine's fault. All of them are production decisions.
02
Early research found learners did better with a human voice than with machine voices, which is the voice principle in Mayer's review. Those machine voices were far worse than today's. More recent work is more nuanced: Chiou, Schroeder and Craig (2020) found that the quality of a virtual human's voice did not significantly affect learning, but did affect how much learners trusted and rated the agent. For production, the implication is that a good synthetic voice is unlikely to harm learning, but poor delivery costs trust, and trust affects whether people take the course seriously.
03
Production fix | Issue
Write them as they should be spoken in the narration script ("خمسة آلاف ريال") | Numbers, currency, dates
Pronunciation dictionary or phonetic spelling per project | Names and brand terms
Add diacritics on ambiguous words in the narration script only | Missing short vowels causing wrong reading
Choose voice and style per role: neutral narrator, warm coach, tense customer | Tone
Match the voice to the audience; review by a speaker from that country | Regional pronunciation
Insert pauses at beat changes; avoid reading lists without breaks | Pacing
04
The EU AI Act (Regulation 2024/1689) includes transparency obligations in Article 50: people must be told when they interact with an AI system, and AI-generated or manipulated content that resembles real people (deepfakes) must be disclosed. These obligations have applied since 2 August 2026 (the AI Omnibus gave providers of generative systems already on the market until 2 December 2026 for machine-readable marking). They matter for courses sold to European clients or EU-funded programmes. Gulf regulators are also issuing AI ethics and deepfake guidance. The simplest compliant practice is also the most trust-building: disclose synthetic voices and presenters at the start of every course.
05
Add a synthetic media note to your scenario script: which voices and presenters are AI-generated, the disclosure line, the pronunciation list, and any consent records. Add it to your module artifact.