Translate audio with an AI dub
Upload a recording, check the translation line by line, and download a dubbed audio track in another language, timed to the original speech.
MP3, M4A, WAV and other audio up to 100 MB and 5 minutes.
Voice sample
Uploaded and processed audio is stored on SoundHalo. Audio access expires after 30 days; expiry does not confirm storage deletion. Delete the stored files here before leaving this page. Speech providers have their own retention policies. Privacy policy.
Example: English to Spanish
A 20-second podcast intro with a quiet music bed, dubbed into Latin American Spanish with the Dora voice and mixed over the lowered original. The transcript was cut into one line per sentence or pause. Each line has until the next one starts, and every Spanish line fit its time without speeding up. Nothing was edited.
| Time | Original | Dub line | Line check |
|---|---|---|---|
0:00.8 → 0:05.1 | Welcome back to the Home Studio podcast. | Bienvenidos de nuevo al podcast Home Studio. | Fits |
0:05.1 → 0:09.0 | Today, clean audio at home. | Hoy, audio limpio en casa. | Fits |
0:09.0 → 0:13.3 | First, pick the quietest room you have. | Primero, elige la habitación más silenciosa que tengas. | Fits |
0:13.3 → 0:17.1 | Soft furniture cuts down on echo, | Los muebles suaves reducen el eco, | Fits |
0:17.1 → 0:20.1 | and always record a short test. | y siempre graba una prueba corta. | Fits |
Original recording
Dub over the lowered original
What you check before anything is spoken
- Every line of the translation sits next to the original speech and its time, so you can read them side by side.
- Lines that look too long for their time, even sped up, are marked before you use any speech characters.
- You can edit a line, translate one line again, or clear it to skip it. The voice is made only when you choose Create dubbed audio.
- After the dub is made, the line check shows which lines fit, which were sped up and which still run over. Listen to the result on the page before you download it.
How the timing works
The original speech is transcribed with the time of every word. The transcript is cut into lines at the end of each sentence, at pauses of a second or more, and before a line passes about 15 seconds. Each translated line starts at the time the original line started, and may use up to 2 seconds of the pause after it. Silence at the edges of each generated line is trimmed, and a line that is longer than the original is sped up without changing the pitch, up to 1.25 times. Beyond that, speech sounds rushed, so the line is flagged instead and you can shorten it.
Translations into Spanish, French or German are often longer than English. The translation is asked to write natural spoken language and to choose shorter wording when a line would run long, which keeps most lines inside their time.
Dub in the original speakers' voices
With a free account, choose "Original speakers' voices (cloned)" as the voice. Transcription then adds speaker labels (up to four people), and each speaker's clearest stretches of speech, up to about 15 seconds, become a reference for their voice. Every translated line is spoken by its own speaker's clone, so a two-person interview stays two voices.
- Before anything is generated, the review shows which voices will be copied and the exact cost of the cloned speech.
- You confirm you have the right to dub the people in the recording in their own voices, and that none of them is under 18.
- People with less than two seconds of speech are not copied; their lines use the first speaker's voice, and the review says so.
- Background music or noise in the reference seconds is copied into the voice, so a clean recording works best. Speech is not separated from music yet.
- Cloned speech is processed by Fish Audio and charged at its real cost, $15 per million bytes of text: up to $0.03 a day on a free account and $0.30 on Pro. The speaker clips are deleted after the dub (if that fails, they expire with the 30-day rule for processed audio).
Example: two speakers in English (both are our own stock synthetic voices, so no real person is cloned here), dubbed into Spanish in their own voices. Each dubbed line scored 0.80 to 0.89 speaker similarity to its own speaker and 0.44 to 0.50 to the other one. The cloned speech cost $0.0051.
Original (English)
Dubbed in the original voices (Spanish)
To use your own voice instead, see voice cloning.
Background music and the original voice
This tool does not separate speech from music or background sound yet. You have two choices:
- Voice only: a track with just the dub, the same length as the original. Use it in your editor over your own music bed.
- Dub over the original: the whole original recording drops by about 15 dB while the dub speaks and comes back between lines. Music stays, but the original voice can still be heard faintly underneath.
Limits
- No account: recordings up to 5 minutes, and a dub preview of up to 300 speech characters (about 20 seconds of speech) and 20 lines. Longer scripts dub the first lines that fit. Speech characters (1,200 a day), translations (5 a day) and transcription minutes are shared by everyone on your network and capped by a site-wide free budget, so they can run out.
- Free account: recordings up to 30 minutes, up to 3,000 speech characters (about 3 minutes of speech) and 200 lines per dub, 20,000 characters a day.
- Pro: recordings up to 30 minutes, up to 20,000 speech characters (about 20 minutes of speech) and 500 lines per dub, 300,000 characters a day.
- Speech characters are shared with the text to speech and voice-over tools. Voices from Giggy count each line as whole blocks of 20 characters. Mixing the dub over the original counts toward the daily cloud processing allowance. Limits reset at midnight UTC.
Synthetic voice and acceptable use
The dub is read by a text to speech voice. It is not the original speaker and not a recording of a real person. Dub recordings you made or have the rights to translate, such as your own podcast, course or voice notes. Do not use it to make someone appear to say something they did not, to impersonate a person, or for spam and fraud. Tell your listeners the dub is AI generated; some platforms require that label.
Related tools
FAQ
Will the dub sound like my voice? +
By default the dub uses a stock synthetic voice you pick, such as Dora or Alex for Spanish, and each voice has a sample to play first. Signed in, you can instead choose original speakers' voices: each person in the recording is cloned from a few seconds of their own speech and speaks the translation. That option is charged at the voice provider's real cost and needs your confirmation that you may dub those people.
Does it remove the original background music? +
No. Separating speech from music and background sound is not available yet. You can download the dub as a voice-only track, or as a mix where the whole original recording is lowered by about 15 dB under the dub. In the mix, the original voice is still faintly audible under the new one.
Which languages can I dub into? +
25 languages: English (US), English (UK), Spanish, French, German, Italian, Portuguese (Brazil), Portuguese (Portugal), Japanese, Chinese (Mandarin), Hindi, Russian, Turkish, Polish, Dutch, Swedish, Danish, Norwegian, Finnish, Romanian, Hungarian, Czech, Indonesian, Vietnamese, Arabic. The original recording can be in any language the speech recognition supports; it is detected automatically.
Can I correct the translation before the voice is made? +
Yes. After transcription and translation you see every line next to the original speech, with its time. Edit any line, ask for a new translation of one line, or clear a line to skip it. Speech characters are only used when you choose Create dubbed audio.
What happens when the translation is longer than the original? +
Each dub line starts at the original line's start time. If the speech is longer than the original line, it is sped up, up to 1.25 times, without changing the pitch. If it still does not fit, it plays in full and the line check shows how far it runs over, so you can shorten it and dub again. Words are never cut. The translation is also asked to choose shorter wording when a line would run long.
What files do I get? +
The dubbed audio as MP3 or WAV (the mix over the original, the voice-only track, or both) and the translated script as an SRT subtitle file with the original timings.
Can I dub a video? +
This tool takes audio files. To dub a video, use the video dubbing tool at /tools/video-dubbing/. It works the same way, returns an MP4 and can keep the original music and background sound.
Where is my recording processed? +
The recording is uploaded to SoundHalo and transcribed by Lemonfox. The transcript is translated by the DeepSeek model through OpenRouter, using only hosts that do not store or train on prompts. Each translated line is read by a text to speech provider: Lemonfox for its languages and Giggy for the others. The recording and the generated audio are stored on SoundHalo so you can download them. Access expires after 30 days, and you can delete the files from the page. When you are signed in, the transcript is also saved to your account library, where you can delete it; the translation is not kept.
What are the limits? +
No account: recordings up to 5 minutes, and a dub preview of up to 300 speech characters (about 20 seconds of speech) and 20 lines. Longer scripts dub the first lines that fit. Speech characters (1,200 a day), translations (5 a day) and transcription minutes are shared by everyone on your network and capped by a site-wide free budget, so they can run out. Free account: recordings up to 30 minutes, up to 3,000 speech characters (about 3 minutes of speech) and 200 lines per dub, 20,000 characters a day. Pro: recordings up to 30 minutes, up to 20,000 speech characters (about 20 minutes of speech) and 500 lines per dub, 300,000 characters a day. Speech characters are shared with the text to speech and voice-over tools. Voices from Giggy count each line as whole blocks of 20 characters. Mixing the dub over the original counts toward the daily cloud processing allowance. Limits reset at midnight UTC.