PG Music Home Forums General Discussion Off-Topic Good evening, I'm looking for software

Good evening, I'm looking for software that, when I upload a song from my computer, will translate the language of the singer in the song into another language that I choose from the list of languages, but in the original voice of the singer himself

I have no first hand experience with it, but I’m told Adobe Firefly will do that. A quick search through my YouTube saves of AI videos on audio.ai says that ElevenLabs AI Dubbing will also do it.

· Core Feature: Translates audio/video to a new language while preserving the original speaker's voice, emotion, and timing.
· Languages: Supports over 20 languages, including Spanish, Japanese, and Polish.
· Best For: This is the most direct match for your request of keeping "the original voice of the singer."

Check out the Eleven labs website demos and see if that is what you need. I’ve gone there and it seems to indicate something different than the video promos. It may be more aimed at speaking voices than singing voices.

I'd suspect that AI can do that. Try some of the AI programs and see if they are able to do what you want.

TL;DR: There's no "one stop shop" for making this happen. It can be done with the help of AI, but it's not simple and will require a lot of work on your part to do it right.

First of all, translation of a song into another language requires an understanding of the meaning of the words, but of the melody as well. Consider:

This is a song (4)
boom boom (2)
I sing it wrong (4)
zoom zoom (2)
But sing along! (4)

Using Google Translate to translate it to Spanish, I get:

Esta es una canción (7)
bum bum (2)
no la cantes mal (5)
zoom zoom (2)
¡pero canta (4)!

I've placed the syllable count after each line to show that the translated version has a different number of syllables than the original. That means the melody needs to be altered to fit the new syllable count.

You'll also note this "raw" translation doesn't keep the same rhyme scheme, as canción, mal and canta don't rhyme.

Google Gemini is "smart" enough to keep the rhyme scheme:

Esta es una canción (7)
bum, bum (2)
La canto sin ton (5)
zum, zum (2)
¡Sigue mi son! (4)

It can be guided to try to match the syllable count:

Esta canción (4)
bum bum (2)
Canto sin ton (4)
zum zum (2)
¡Si-gue mi son! (4)

Although obviously the translation isn't literal, as "I sing it wrong" doesn't necessarily mean "I sing off-key":

This song
boom boom
I sing off-key
zoom zoom
Follow my rhythm!

This is why translation is the hardest bit - you'll need to have something that captures the meaning of the lyric, even if it's not a literal translation. After telling Gemini I wanted the lyric to mean "incorrectly" instead of "off-key", Gemini offers this translation after a lot of thought:

Es un cantar
bum bum
Lo canto mal
zum zum
¡Ven a cantar!

Gemini offered the following notes:


But you can see that it's rhymed "cantar" with "cantar".

All of which is to say that translating while keeping the rhyme scheme and syllable count isn't trivial, but will require interaction to verify the correctness of the result. It's an iterative process, and is likely to exhaust what you can do free with an AI pretty quickly. You'll likely have to as AI to find phrases that have similar meaning in the target language, and try using them instead of the original lyric.

Once you have a translation, you then need to get it sung in the new language. Perhaps the simplest way to do this is to create a "cover" in Suno. This requires that Suno doesn't see this as a copyright song. Of course, you're doing this with a song you've got full rights to.

Once you do that, you type new lyrics into Suno, and it will create "cover" the song with the new lyrics. However, it's going to us a different voice instead of the original when creating the "cover". You'll need to describe the voice of the singer to Suno so that the voice it uses will be similar in character to the original singer.

So you'll end up with song singing the new lyrics that you've translated into your target language.

You'll then need to extract the vocal stems of the song from Suno, and convert the vocal into the original singer's voice (but in the new language).

Now would be a good time to mention that harmony vocals will need to be done in a separate pass!

There are tools like RVC, but Controlla Voice seems to be the best option at the moment. You'll need to upload audio to train it with your singer's voice. The training material should match the style of the target song. For example, if you train using soft vocals and try to apply it to a hard rock song, it's not going to work.

Once that's done, you can upload the vocal stem from Suno into Controlla Voice (or some similar tool) and it will apply your original singer's voice to it. It will be in the new language, in the style and performance of the Suno voice, but with your singer's timbre applied.

If you've done all this work, you'll also want a native speaker to listen to the song (especially the Suno version) to make sure it sounds right. Different language have different rhythmic rules, and what sounds good in one language may sound terrible in another. Suno is pretty good at automatically making these adjustments, though.

To summarize, it's a multi-step process, using different tools, and a lot of work is required for it to turn out well:

1. You'll need to translate the song into the new language. If you aren't fluent in that language, you could use an AI program like Google Gemini to perform the translation, using Google Translate to make sure the translation close to the intended meaning. This would be an interactive process.

2. You'll then need to covert the vocal into the new language. You can do this with a tool like Suno, where you load the song, and then create a "cover", giving it the new lyrics. You'll then need to extract the vocal stems to get the isolated vocal.

3. You'll then need to convert the Suno vocal back to your singer's voice, in the target language. You can use a tool like Controlla Voice or some other RVC tool that's been trained with samples of your singer's own voice. It will apply your singer's voice while keeping the native accent of the Suno vocal.

Of course, this isn't how the pros do it. For the Spanish version of "Rocking Around the Christmas Tree" done in Brenda Lee's voice, the producers hired a native Spanish singer, and then used an RVC-like tool trained using Brenda Lee's voice to covert the vocals. That's the best way, as the singer could serve as another set of eyes to make sure the translation made sense. But you'd have to make sure they'd know they're vocals would be substituted with someone else's voice!

It can help to know what contractions are posssible in the language. Sometimes one can drop a sylable with the contraction.

Edit: Also the vernacular, e.g., "Ain't gonna do that no more".

That made my head spin a few turns, their brother David. Fascinating commentary on the subject. It may have caused the OP to run for the hills!!

I am at about 1000 hours into the learning curve with AI. None of my LLMs are suited to tackle that issue. I now have an excellent medical LLM that only searches my data and nothing else. Several million pages of first-class peer-reviewed data, like The Merck Manual.
The issue may be solvable with AI, but that is just a guess. My best guess is that public AI would be woefully unsuitable for the task.

Happy New Year to everyone. I am wishing for the best for all of you for 2026!

Billy