Conch voice domestic launch, may be the best Chinese voice dubbing products

There has not been a domestic production for the content of the production of excellent voice-over products, either you can only use the API or the product is okay sound modeling does not work.

For example, overseas ElevenLabs Although the English language is okay but the Chinese language is really pulling across, the main problem of the open source model is that the model quality is relatively poor, specifically in the poor sound quality, there is a clear sense of current, multi-language mixed scenes and multi-sound and multi-meaning word expression is inaccurate and so on.

MiniMax went live a month ago. Conch AI international version, online second voice cloning function, Chinese and English read aloud very well!The voice generation feature went live on the Domestic page a few days ago, and I was really surprised when I tried it out.Supports dubbing in more than 17 languages, precise control of various emotional expressions, and hundreds of sound libraries to meet different needs.The

The best part is that the audio quality is really high, with none of the common problems with sound generation, and highly customizable options that make it different when tweaked, even though the same tones are used.

Let's take a look at this little slice of what I did, and then we'll look at the effectiveness of Conch's speech generation capabilities, both in terms of functionality and effectiveness.

Detailed feature customization options

Let's take a look at the features of Conch Voice, it's really powerful and detailed, Conch has a huge sound library supporting more than 17 languages.Each language supports a wide range of tones, and you can freely choose between male and female voices, as well as ages.The

You can filter to find tones of any identity and age background you need, for example, our video script needs an elderly man with a sense of justice, which can be quickly accessed through this filter.

It is also possible to customize the tone in great detail after selecting it.

First of all, you can customize the emotion of the tone, including happy, sad, angry and other five or six kinds of emotions, for example, the following audio I show the warm girlfriends of this tone of different emotional performance, you can clearly hear, very natural.

Then is the speed of speech, this is very good to understand the higher the value of the faster he speaks, the volume is also very good to understand, the higher the value of the voice of the higher voice, the tone of this I tried to roughly can be understood as the value of the larger the voice of the more pointed, the smaller the value of the voice of the more mellow.

By controlling these four customization options, weCan be tuned to sound very different, even if you pick the same tone, really fun to try out!The

Another tip is to add '' where you need to add a pause, this kind of markers can let the model generate the sound pause for a specified time, if you have this kind of need you can try it, the conch itself can also judge the pause needed.

Powerful generation effects

In addition to the rich timbre in addition to the model of the conch ontology is also very powerful, we have used some open source TTS know that many of the most common problems with the model is the sound quality problems, there is a sense of current, some distortion some of which is intentionally added some of which is a problem of training.

Here I found a relatively long paragraph I wrote some time ago for him to generate a bit of dictation thatYou can listen to the sound quality which is very good, and the pauses are natural, and he emphasizes when he needs to emphasize.The

Another common problem with speech modeling is the generation of ultra-long content.Many models support very short text lengths, Conch supports up to 10,000 characters., basically that's the length of a longer manuscript and a chapter of a novel, and it's perfectly adequate.

The following is that I found a period of time ago Wu Enda two thousand words of the manuscript he read very well and no problem, the generation of speed is also very fast, he can be generated while previewing, very time-saving!

The last trickiest problem is the multi-language mixed scene and polyphonic scene, some relatively good speech models also often have problems, specifically let the AI generated a test text, a paragraph contains five different languages, the conch reads perfectly, this is really too strong.

Good morning! Je suis très heureux de vous rencontrer. Ich liebe Musik und Kunst. It makes me very happy. ¡Buenos días amigos! Let's go to lunch.

This is a polyphonic scene, and he accurately determines the pronunciation of the different positions of "walk a line" (háng), "first" (xíng), "bank" (háng), "trip" (xíng), and handles the very complex polyphonic utterances very well.

I'm going to get a head start today and stop when I go one way. I have to go to the bank tomorrow, so if I get stuck in a traffic jam, my trip may be affected.

That's all for the introduction you can play around with it more and use it in these following places:

Conch Voice: https://hailuoai.com/audioHailuo

Domestic API service: https://platform.minimaxi.com/document/T2AV2

Last year I still always talk to my friends about when the country can also have a powerful voice-over product like ElevenLabs, now we have obtained even better results than ElevenLabs, 24 years we have come up with a model that rivals the first-class level from picture to video to audio, and I hope that this year, domestic AI vendors can give us more surprises.