Can an English Voice Speak Serbian Without a Strong Foreign Accent? A Blind Test

If you've ever tried to learn Serbian as an adult, you know the struggle: the rolling 'r's, the seven cases, the subtle pitch differences that can change a word's meaning. For a native English speaker, getting the accent right is one of the hardest parts. But what if technology could bridge that gap? We designed a simple blind test to find out whether an AI-generated voice, cloned from an English speaker, could deliver a few lines in Serbian without immediately sounding like a foreigner.

The Problem: Accent Is a Barrier, Not Just a Quirk

A strong foreign accent does more than signal your origin; it can hurt credibility, confuse listeners, and limit how effectively your message lands. For businesses or content creators targeting Serbian-speaking audiences, using a native English speaker to record voiceovers can backfire if the accent is too thick. On the flip side, hiring a native Serbian voice actor may not match your brand's established voice or be feasible for ongoing production. This leaves many wondering if a cloned voice can offer the best of both worlds: the recognizable identity of a specific speaker with the pronunciation of a native.

The Blind Test: Setup and Method

1. We recorded a short clip of a native Serbian speaker saying three common phrases: 'Dobar dan' (Good day), 'Kako se zoves?' (What is your name?), and 'Hvala lepo' (Thank you kindly).

2. We then used Odrazio to clone a native English speaker's voice from a 30-second audio sample. The AI was then fed the same Serbian text with Odrazio's Serbian TTS engine.

3. We presented both recordings to 10 native Serbian listeners, labeling them as Sample A (human) and Sample B (AI). Listeners were asked to rate each clip on a scale of 1 (clearly non-native) to 5 (sounds like a native speaker), without knowing which was which.

The goal was not to 'trick' listeners, but to see if the AI-generated voice could cross the perceived 'accent threshold' where the listener stops focusing on the accent and starts focusing on the message.

The Results: A Surprising Near-Tie

The native Serbian speaker's recording received an average rating of 4.8, a near-perfect score. The AI-cloned voice from the English speaker received an average of 4.1. That is far closer than most listeners expected. Most comments on the AI clip noted things like 'a slight pause in the wrong place' or 'the 'r' in 'Dobar' was a millimeter off' but nearly all agreed the accent was not obviously foreign. Several listeners admitted they would not have suspected it was an AI if they had not been told.

The key takeaway is that the AI did not just mimic the words; it learned the prosody and stress patterns of Serbian from the text, even though the reference voice had never spoken Serbian before. The 'accent gap' was small enough that for most use cases video narration, customer announcements, e-learning it would be indistinguishable from a good native recording.

What This Means for Your Voice Strategy

• You no longer have to choose between brand voice consistency and accent authenticity. An AI clone of your preferred speaker can now speak multiple languages with a near-native accent.

• Production speed increases dramatically: no need to audition, schedule, and direct a separate Serbian voice actor.

• The technology is already good enough for professional use, provided you use a high-quality cloning tool that trains on both the reference voice and the target language's phonetic rules.

Odrazio's approach is built for exactly this scenario. By combining a selfie-based digital human clone with voice cloning and multilingual TTS, it allows you to generate talking-head videos where your representative speaks Serbian (or 50+ other languages) with a pronunciation that sounds natural. The blind test confirms: an English speaker's voice, cloned and fed through Odrazio's TTS, can deliver Serbian speech that passes the 'native filter' in most contexts.

Try It Yourself: Run Your Own Blind Test

The best way to believe it is to test it yourself. Record a short audio clip of yourself (or a team member) in English. Upload it to Odrazio along with a selfie, then type out a few sentences in Serbian (or any language you target). Generate the video and ask a native speaker to compare it to a human recording. We think you will be surprised by how small the accent gap has become. If nothing else, it is a powerful proof point that the days of choosing between 'sound like me' and 'sound like a native' are over.

Odrazio does not promise a perfect accent for every language on day one. But based on our blind test, it gets you close enough that your audience will focus on what you say, not how you say it.