AI voice or a real voice actor for your commercial? When it really pays off (and what the AI Act says)

A synthetic voice can now be generated from a script in a few minutes, and for many videos it is a sensible choice. For others it is not. The difference lies less in the timbre, which the best services reproduce well, than in what the script asks of whoever reads it: the more interpretation it needs, the smaller the advantage of an AI voice.

The two routes

• AI voice: you pick a voice in a speech synthesis service (ElevenLabs, Artlist and others), paste in the script and generate the reads, from the web interface or through the API.
• Voice actor: a person reads the script, from home or in a studio, starting from a brief.

The two routes often coexist in the same project: an AI voice as a temp track on the edit, a voice actor for the final version.

Where an AI voice works well

• Informational scripts where the read needs to be clear more than expressive: tutorials, e-learning, internal videos, instructions.
• Social content with tight deadlines and a short life.
• English above all: it is the language on which the models are most mature. In Italian the results are good but less consistent, especially on cadence and accents.
• Multilingual versions of the same informational video, when the alternative would be a casting for each language.

Where you hear the difference

The weak points of a synthetic voice are the ones that require understanding the script, not just pronouncing it:

• emotion and subtext: a line said with a smile, a pause that carries weight;
• irony and comic timing, where the timing of every syllable matters;
• proper names and brands, which the model may mispronounce or pronounce inconsistently;
• dialogue between two voices, the hardest thing to get right convincingly.

Then there is file quality. On the basic plans many services only export MP3, and the compression is audible: in the final mix, next to music and effects, a compressed voice loses presence. WAV is usually reserved for higher plans.

With experience of the tools and good editing, high-quality results are still achievable: you generate many variants, play with punctuation to push the model towards a different intention, then edit the best pieces together, sometimes syllable by syllable. It is a work of trial and collage. That is why the promise of speed holds for simple scripts: when the ambition rises, the editing time can cost more than a session with a voice actor.

How the work changes

With an AI voice you start by choosing the voice, generate the reads, fix the parts that do not work and edit. Client revisions to the script are quick: a changed line is regenerated without calling anyone back.

With a voice actor, today the recording almost always happens remotely: the voice actor receives the brief, records from home and sends one or more reads, then the corrections are discussed. A studio session with live direction is the slowest route and, today, the rarer one, chosen when the performance is the heart of the project.

A practical detail about home recordings: every voice actor has their own setup and often applies processing to the file (compression, EQ, noise reduction). Whoever then mixes the video is better off also receiving the clean, unprocessed recording, so the sound of the voice is decided in the mix together with everything else.

Which one to choose

• TV and radio commercials, branded content, documentaries, emotional content: a voice actor. That is where understanding of the script and performance make the difference, and you can hear it.
• Social media, informational videos, e-learning, internal videos: an AI voice is a real option, especially in English and when the script does not call for subtle acting.
• In-between cases: you can start with an AI voice as a temp track and decide on the edit whether it is enough or a voice actor is needed.

Rights, licences and the AI Act

The service licence. Commercial use of an AI voice depends on the plan: with ElevenLabs, for example, content generated on the free plan cannot be used in a commercial project, while content generated on a paid plan can. It is worth checking the terms of the service you use before delivery.

Cloned voices. Cloning the voice of a real person requires their explicit consent, and the services require it in their terms of use. This also applies to the voice of a spokesperson or of a company employee.

The AI Act. From 2 August 2026 the transparency obligations of Article 50 of the European regulation on artificial intelligence apply:

• whoever provides the speech synthesis service must mark the generated audio in a machine-readable way, so that it can be detected as artificial (for systems already on the market before that date, the deadline is 2 December 2026);
• whoever uses the system must disclose deep fakes, meaning generated or manipulated content that resembles real people, places or events and could appear authentic. A cloned voice of an existing person falls here. In works that are evidently artistic, creative, satirical or fictional the obligation is lighter: it is enough to disclose it in a way that does not hamper the enjoyment of the work.

For a generic synthetic voice that imitates no one, the case is less clear-cut: the European Commission guidelines on transparency obligations help to assess it. When in doubt, disclosing it costs little. For a specific project the final assessment belongs to a lawyer.

What you need to get started

The voice casting, human or AI, can be done by whoever handles the video's audio, together with editing, mixing and creative direction. A brief is enough to start:

1. what kind of voice: male or female, young or mature;
2. the mood: corporate, warm, energetic, ironic;
3. where the video will be released;
4. the reference video or, for radio, the programme or the product.

Write to me with the brief, and we will choose the right voice for the project together.

Sources

Regulation (EU) 2024/1689 on artificial intelligence (AI Act), Article 3 point 60 and Article 50 (eur-lex.europa.eu) · European Commission, "Guidelines on transparency obligations for providers and deployers of certain AI systems" (digital-strategy.ec.europa.eu) · ElevenLabs, Terms of Service and Prohibited Use Policy (elevenlabs.io).