AI voice & dubbing

Choosing an AI Voice When the Dub Copies the Speaker

Choosing an AI voice for a dubbed video starts with one question: should viewers hear the original speaker? In mydubly that is how every dub works. Each person's translated speech is a copy of their own voice, taken automatically from a short, clean sample of that person in your file, so there is no voice list to browse. The real choices move to the recording: clean, close-mic speech copies best, people taking turns keeps each voice clear, and someone with too little clean speech gets the copied voice of the person they sound most like, or a stock voice matching theirs.

8 min read · Updated

Start with what the video is for

mydubly text to audio voice ryan clear with style sample card

Before uploading anything, write down in one line what the video must achieve. "Teach a nurse how to set up an infusion pump" and "get scrollers to stop on our product launch" are different jobs. The first needs clear diction for a viewer who is concentrating; the second needs energy in the opening seconds.

Then ask whether the speaker's identity is part of the job. For a presenter-led channel, a course taught by a named expert or a founder's update, hearing the same voice in the new language keeps continuity, and that is what a copied voice provides. For a screen recording narrated by someone viewers never see, the voice still comes from that narrator, so the recording quality of the narration decides how the dub sounds.

How the copied voice suits each content type

mydubly text to audio voice aiden bright with style sample card

Because the dubbed voice comes from the recording, each content type has its own strengths and weak spots.

Tutorials and software walkthroughs
Usually one narrator recorded close to the mic, which gives the copy a clean sample to work from.
Lectures and courses
A clip-on or headset mic copies well; a distant room mic with echo gives a weaker sample.
Marketing and short-form social clips
Often music-heavy, so check that the clip contains some clear speech; with too little, a stock voice matching the speaker is used.
Corporate and internal messages
A single presenter at a desk is close to ideal for copying.
Documentary-style storytelling
A clean voice-over narration copies well; each interviewee is read in a copy of their own voice.
Meditation, wellness and sensitive topics
Soft speech over ambient music needs some stretches where the voice is clearly audible.

Audience, register and expectations

mydubly text to audio voice aiden energetic with style sample card

Think about who is listening and in which language. The copied voice keeps the speaker's character, meaning timbre, pitch and how they sound, while pronunciation is natural for the target language; the original accent stays with the original language. For Spanish and Portuguese, the language picker also offers regional variants, which shape the translation's wording; the voice itself has no regional accent setting. If you have viewers or colleagues in the target market, ask them whether the dub sounds natural for the genre.

Consider continuity with the picture too. In talking-head videos, a copied voice naturally fits the person on screen. When several people appear, each is voiced in a copy of their own voice, so viewers can follow who is talking by ear, as in the original.

Why pace and density still matter

mydubly text to audio voice ryan angry with style sample card

Translations are often longer than the original, and the dub has to fit into the time the original speaker used. A fast talker leaves few pauses, so a longer translation of their speech needs more speed-up, and the delivery can start to sound hurried even in a voice that sounds like theirs.

In mydubly, small speed-ups up to about 1.08× are treated as inaudible, and the tempo can rise to about 1.15× by default when a line needs more room. If you are recording for translation, short pauses between ideas give each line room to fit. If the original is relaxed and pause-filled, timing pressure rarely shows. The timing explainer details how lines are placed.

How mydubly decides the voice for you

mydubly text to audio voice ryan bedtime with style sample card

mydubly has no voice picker, voice list or voice previews. When you translate a file with full output on, it works out how many people speak (up to 8) and which lines each says, takes a short, clean sample of each person, up to about 10 seconds, with music and background noise removed, and speaks each person's lines in a copy of their voice in any of 19 languages: all 21 supported languages except Czech and Vietnamese, which get subtitles and translated transcripts.

  • A voice per person: each speaker keeps their own copied voice, and each line follows the pace and energy of the original delivery. Transcripts do not label speakers.
  • Natural pronunciation: the voice character is copied, and the words are pronounced the way the target language expects.
  • Automatic fallback: someone with too little clean speech to copy, under about 3 seconds, is voiced with the copied voice of the person they sound most like, if they sound alike; otherwise, for example in a music-heavy clip with very little talking, a stock voice matching theirs (female or male) is used.
  • Privacy: the video never uploads; only the audio is sent, over HTTPS, and the voice sample is deleted with the rest of the job's files 30 minutes after the job ends or after you last use its editor.

Only upload voices you have the right to use, such as your own or those of people who have agreed to a translated version. For the technical background on how a model speaks in a voice taken from a short sample, see Chatterbox TTS. The AI dubbing page lists the languages and output files.

Testing the copied voice on a short clip

mydubly text to audio voice ryan horror with style sample card

Testing on your own footage beats guessing, because your recording, pace and subject are what the copy has to handle.

  1. Cut a 2-minute clip from your video that includes a clean stretch of each speaker, a name, a number, a question and the fastest stretch of speech.
  2. Run the clip in your target language.
  3. Listen for whether the dubbed voice sounds like the speaker and whether the pronunciation sounds natural in the target language.
  4. If the voice sounds rough or roomy, look at the source: noise, echo or music under the speech weaken the sample.
  5. If you hear a stock voice, the clip had too little clean speech to copy; try a section with more clear talking.
  6. If you can, ask a native speaker of the target language to watch the result with the picture and note anything that sounds off.
Example: testing the copy

Suppose a cooking channel wants Spanish (Latin America) versions of its 12-minute recipe videos. The host talks fast over a sizzling pan and background music. The team's first 2-minute test clip comes from the busiest part of the cook and has very little clean speech, so the run uses a stock voice. A second clip, taken from the host's introduction at the counter, gives a clear sample, and the Spanish dub carries the host's voice. Each test is 100 credits, so both cost 200 credits, or $0.20. The team starts recording with a clip-on mic so every episode has clear stretches of speech, and each full episode then costs 600 credits ($0.60).

Keeping one voice across a series

Viewers form an association between a channel or course and its voice, and a copied voice extends that association into every language you dub into. Because the voice is taken from each file, consistency comes from consistent recording: the same presenter, a similar mic and a quiet room for every episode.

Listen to the opening of each new dub before publishing. An episode that was mostly music, or recorded in a noisy location, may fall back to a stock voice and sound different from the rest of the series. Keep a simple note of which episodes you have checked in each language.

Mistakes to avoid and limits of the copied voice

Some problems no voice can fix, and some recordings create problems that look like voice quality issues.

  • Testing on a clip with almost no clean speech, then judging the copy by a stock-voice result.
  • Recording far from the mic or in an echoey room; clean, close-mic speech copies best.
  • Expecting the speaker's original accent in the new language; pronunciation follows the target language, and the voice character is what carries over.
  • Expecting flawless speaker separation in heavy crosstalk; people talking over each other, very short interjections or two very similar voices can land in the wrong voice, so subtitles may serve those panels better.
  • Expecting a voice to fix a weak translation; meaning errors come from the text, so review the translated transcript.
  • Uploading someone else's voice without the right to use it.

Where to go next

Cut a 2-minute test clip with a clean stretch of each speaker and run it on AI dubbing. Once it sounds right, our guide on how to dub a video with AI walks through the full run. If your videos feature several speakers, read single voice vs multi-voice dubbing and translating videos with multiple speakers before committing.

Frequently asked questions

Can I choose a male or female AI voice in mydubly?

There is no voice picker. Each person is dubbed in a copy of their own voice, so the dub already matches the people in the recording. Someone with too little clean speech to copy gets the copied voice of the person they sound most like, if they sound alike; otherwise mydubly uses a stock voice matching theirs, female or male, automatically.

Will my voice sound the same in every language?

Each language run copies the same people from the same recording, so the voice character carries across your language versions, while pronunciation follows each target language. How close the copy sounds depends on the recording, so check a short clip in each language before translating a whole series.

Does the voice affect translation quality?

No. The translated text is produced before the voice is applied, so the words are the same whichever voice reads them. The voice affects delivery, pace and how well emphasis lands, while meaning comes from the text. To improve meaning, review the translated transcript.

What should my test clip include?

Use a stretch of about 2 minutes that contains a clean passage of each speaker plus the hardest parts of your content: a name, a number, a question and the fastest speech. That shows how the copy sounds and how it handles pronunciation, intonation and timing pressure. Since mydubly's minimum for a voiced run is 2 minutes, a clip that length costs 100 credits.

Can I choose a regional accent for Spanish or Portuguese?

The language picker offers Spanish (Latin America) and Spanish (Spain), and Portuguese (Brazil) and Portuguese (Portugal). The variant guides the translation's vocabulary, grammar and spelling, which the voice then speaks. The voice is the speaker's copied voice with the voice model's standard pronunciation for the language, and there is no regional accent setting, so test a short clip with your audience. See translating video for regional audiences.