Why the translated voice sounds robotic, too fast or not like you

This article is about the voiceover in Translate Audio, with a separate section for Translate Video. All voiceover options are changed in the settings: tap the settings icon (sliders) in the top right corner and open the Voiceover tab.

Quick check: symptom, cause, fix

Most complaints come from two things: the translation was read by a standard voice, or the voice reads the translated text at its own pace without following the original.

If the voice sounds robotic or flat

A robotic sound usually means the translation was read by a standard ElevenLabs or OpenAI voice. These voices are the same for everyone and don't imitate the speaker. The OpenAI voice is neutral by design.

• Open the Voiceover tab and choose an option under With cloning: MiniMax (the app describes it as the most accurate in timbre and emotion) or Qwen 3. Then tap play again.

• For the most natural result choose ElevenLabs Dubbing v2. It dubs the original recording itself, keeping its pace and timbre. It costs 4.4 tokens per second, so a one-minute message costs about 264 tokens.

• A clone made from a recording with music, street noise or several voices also sounds unnatural. Use a recording where one person speaks clearly in the first minute.

If the translation is too fast or too slow

The voiceover reads the translated text at the voice's own pace. It doesn't stretch or shrink the speech to the length of the original, so a translation can end sooner or later than the original.

• MiniMax: in the MiniMax block open Settings and move Speed below 1.0x. The range is 0.5x to 2.0x.

• Qwen 3: choose Slow in Speech style. Speech style works only with Clone from file, not with a saved voice.

• ElevenLabs and OpenAI standard voices have no speed setting in Translate Audio.

• To follow the tempo of the original, choose ElevenLabs Dubbing v2.

After changing speed or style, tap play again. With MiniMax the repeat voiceover of the same result doesn't charge the 180 tokens for cloning again.

If it doesn't sound like you or the speaker

Clone from file copies the voice of whoever speaks in the recording. If the result doesn't sound like that person, check these causes in order.

1. A standard voice was used. This happens if you answered Use AI voice in the Clone voice? dialog (with Don't ask again the app keeps doing it), chose ElevenLabs or OpenAI in the settings, or picked a language that no cloning option supports. Choose MiniMax or Qwen 3 on the Voiceover tab.

2. The recording was too short: Qwen 3 needs at least 3 seconds, MiniMax at least 10. Then the app uses a standard voice and says «Audio too short for voice cloning, standard voice was used».

3. Cloning failed and the app said «Voice cloning is unavailable right now. A standard voice was used.» Tap play again later or choose MiniMax.

4. A saved voice was selected instead of the recording. If My MiniMax voice or My Qwen voice is active, the translation is read by that saved voice. Select Clone from file.

5. The first 60 seconds are noisy. MiniMax and Qwen 3 take the voice only from the first minute of the file. Music, another person or loud background there make the clone less similar.

If you want your voice on someone else's message, for example to reply, save your voice once in Settings → My Voices and select it as My MiniMax voice or My Qwen voice. Tips for a good sample are in Tips for the best voice clone quality.

If the accent or dialect is wrong

Translate Audio can't set an accent or a regional variant. The language list has one entry per language: Portuguese without a Brazil or Portugal choice, English without US or UK.

• A clone keeps the timbre of the speaker and often their own accent. A person who speaks English will usually sound like a foreigner in Portuguese, even with a correct translation.

• The standard voice is the same fixed voice for every language, and you can't pick another one in Translate Audio.

What you can do:

• Copy the translation, open Text to Speech and choose an ElevenLabs voice with the filters by language and accent. The voice will be someone else's, but with the pronunciation you need.

• For a video, use Translate Video: its language list has regional variants such as Portuguese (Brazil) and English (United States).

If the emotions are lost

MiniMax and Qwen 3 copy the timbre of the voice, but they voice the translated text, so the intonation of each phrase is not taken from the original. The model chooses it from the text unless you set it.

• MiniMax: set Emotion to Happy, Sad, Angry, Fearful, Surprised, Calm or Disgusted. One emotion applies to the whole translation. Auto (recommended) lets the model decide.

• Qwen 3: Speech style Calm, Energetic, Friendly, Soft or Professional.

• ElevenLabs Dubbing v2 works from the original recording and stays closest to its delivery.

If two people speak in one voice

In Translate Audio, every option except Dubbing v2 reads the whole translation with one voice. MiniMax and Qwen 3 make a single clone from the start of the file, so in a conversation both people get the same voice, made from the first minute of the recording.

To give each speaker their own voice, choose ElevenLabs Dubbing v2. It detects the speakers and clones each of them.

If the pauses are not kept

MiniMax, Qwen 3 and the standard voices read the translation as one continuous text, so long pauses of the original disappear. Translate Audio has no setting to keep them. ElevenLabs Dubbing v2 dubs the original timeline and keeps the timing.

In Translate Video

Translate Video has its own settings: tap the settings icon in the top right corner of Translate Video before you start.

• One person was dubbed with two voices, for example male and female. With HeyGen and Number of speakers set to Auto, HeyGen decides how many people speak and can find two. Set Number of speakers to 1 and translate again.

• The voice sounds robotic or far from the original: with HeyGen turn on Enhanced cloning (twice the price), or choose ElevenLabs, which clones each speaker and keeps their timing. If the original is noisy, turn on Enhance speech quality.

• Wrong dialect: choose the regional variant in the language list, for example Portuguese (Brazil).

All Translate Video options are described in How Translate Video works.

When to contact support

If none of this helped, tap Not happy with the result? in the Feedback block under the result and describe what you hear. The form attaches the job, so support can listen to the same recording. The general form is at vocalingo.app/support.