What is an AI Singing Voice?
Similar to AI Voice Cloning solutions that can turn a snippet of text into uncanny human-sounding speech, AI Singing voices are designed to generate realistic singing voices for music. There are two main types of AI Singing Voice technologies being used today:
Singer Voice Conversion (SVC)
Singing Voice Conversion is the same technology that is being used to generate AI Covers of popular songs like "Eric Cartman sings Love Yourself" or "Drake sings Munch by Ice Spice". You can think of SVC technology like a voice-changer for singing. It takes a reference or input vocal of one person singing, and transforms the character and accent of their voice to sound exactly like someone else sang it with the same rhythm, melody, and vibrato as the original singer. As artists, we strongly support the development of SVC technologies as it directly opens up new opportunities for singers to sing on behalf of other artists, brands, characters, or even politicians. Controlla has been a pioneer in creating these opportunities - any selfless and talented singer can apply to help the world sing.
Singing Voice Synthesis (SVS) SVS technologies, on the other hand, aim to democratize the creation of singing performances in general. Many of them let music creators input lyrics in combination with a melody to generate full singing performances in various voices. The earliest creators of these plugins or software tools are Vocaloid or Synthesizer V. They've gotten pretty realistic in the past few years but still don't carry the same realism, emotion and quality that Singer Voice Conversion tools can offer.
Now that you know the core technologies and functionalities of AI Singing Voices, let's take a look at some of the best products and services to take advantage of as a music creator, singer, producer, mixing engineer, or artist.
1. Controlla Voice - Highest quality, ethically trained (SVC)
ControllaVoice stands out for its unwavering commitment to ethical standards, ensuring transparency in data sourcing and staunch user data protection. While they won't help you steal celebrity voices, instead they aim to help you unleash the full potential of your own voice. Controlla believes every artist, character, person or politician will have their own AI singing voices which they use to build their brand and express themselves through singing. ControllaVoice models were designed with flexibility in mind to both accurately model the world's best singers while helping non-singers sing anything in any language. Offering unparalleled quality at the lowest custom voice prices, this is the go-to for both amateurs and professional music creators if they're looking to generate vocals in their own voice. It's still in early access and thousands of artists have already been blown away by the results they can create using the power of real singers and AI SVC technologies. You can even apply to earn passive income from your own singing. ControllaVoice has already been used by major labels for multilingual songs and notable producers like Jay Stolar from Hume Collective where it was even featured on CNN to simulate their interviewer singing in their track live.
Lowest price for training your own AI singing voice
Works in any language
Auto-upgrading voices that improve over time
Works for non-singers (you can train a model on speech that generates singing)
Highest output quality - accepts over an hour of vocals for custom voices
All underlying models were trained on ethically sourced and commercially licensed vocals
No unauthorized celebrity or character voices.
The Founder is kind of a narc when it comes to the ethics and morals of AI singing voices. He even spoke out against voice theft at the US Copyright Office's AI Music Panel.
Price - $30/mo for unlimited singing ($10 to try)
2. Synthesizer V - Create vocal music from nothing (SVS)
SynthV is an amazing, realistic sounding option for Singing Voice Synthesis. You can draw out melodies the same way you would on a piano roll or MIDI editor. For each note you can type words or syllables and create brand new lyrics and vocals on the fly. The caveat is that this only works for the voices they sell, which costs around $100 for the best voices. If you want to use generate vocals like this in your own voice, it will often require many hours of data and there aren't any full solutions on the market for you to do this in an affordable way. You can however, combine the power of SynthV to draw any lyrics and melodies in one of their voices, then use ControllaVoice to convert that performance into your own voice. This is currently the quickest, most affordable way for you to start using Singing Voice Synthesis technologies on your own voice.
Works in English, Japanese and Chinese
Real-time feedback when editing melodies/words
Highest quality singing voice synthesis options
Easy to use interface, runs locally so no subscription
Vocal presets are expensive
Can't train a custom voice unless you combine it with ControllaVoice
Results are audibly synthetic if you don't find the right words/melodies that fit together
Price - $90 one time, extra voices cost around $90 each
3. AI Hub Discord Server and Open Source tools (SVC and Text To Speech)
The AI Hub Discord server is a testament to the strength and innovation of the open-source community. This was where Heart on my Sleeve came from, and it's also the source of many unauthorized AI covers across social media. It's a hub for resources, tools, and cutting-edge developments in AI voice technology, backed by a relentlessly adventurous and unique community.
Ironically, it also started as one of the biggest voice theft platforms. They built out the infrastructure to let anyone clone celebrity voices and share them in an open source way using their discord moderators and bots. After tons of legal backlash from majors labels and the RIAA, it has simply become a valuable source of information for those looking to train their own voices for free, without giving up data to corporations.
You'll find tons of guides in AI Hub for using open source models like Diff-SVC, So-vits-SVC, RVC, and DDSP-SVC. Be careful using their Colab notebooks, however, as all samples may be stored on their servers. This is a good option for learning how to train your voice with open source code, but make sure you use your own hardware for training and don't give up samples of your voice to publicly hosted scripts or notebooks.
Can run locally if you have a powerful GPU
Lots of bad advice from beginners
Most tools they recommend are tailored towards stealing voices
Can't use commercially as their models are trained on unlicensed data
Price - Free
4. VoiceMod - Text To Sing
VoiceMod, while celebrated among gamers for its transformative capabilities, also ventures into the realm of singing with their latest Text To Song tool. With a simple lyric prompt, voice selection, and instrumental selection, it will output a full song with synthetic vocals singing the lyrics you provide. It's still in it's early phases and doesn't seem to output anything worth a Grammy yet, but you can download the vocal stems and remix them as you see fit. Many of the melodies are preset for now but in the future you will be able to input your own lyrics and melodies to generate any vocals you imagine. The voices they provide are really good and almost sound real with the right words. Price - Free to try
5. UberDuck - a pioneer in AI generated raps
UberDuck was one of the first AI Rap generators allowing anyone to generate raps and speech in celebrity voices like Drake, Eminem, Morgan Freeman, and more. It's unclear who owns the content generated with an unlicensed voice on this service. However, it's a fun meme making tool if you're looking for a fun way to experiment with AI voices. Their singing voices specifically don't have the highest quality for creating music, but can be an invaluable tool in creating memes for social media to build your following.
Generate full rap verses with instrumentals from prompts
Free to use meme voices
Rap/Singing quality is low
Only one custom voice for subscription
Price - $10/mo for limited credits and one voice
Conclusion: What's right for you?
There's tons of solutions available to train your own AI singing voice. The first question you'll want to ask yourself is what do you want to create?
Are you looking to make memes and AI covers with celebrity voices? UberDuck is one of the longest standing option as most of the other platforms are taken down due to legal issues. Are you looking to make AI covers in your own voice to expand your audience? Or do you want to create your own AI singing voice to enhance your music and style? ControllaVoice is the best ethically trained option for tapping into the true potential of your own voice in a commercially viable way. You can try ControllaVoice for as low as $10 during early access
Are there any AI Singing Voice VSTs or plugins in a DAW?
Not yet. ControllaVoice is currently developing an AI Singer Voice Conversion VST slated to launch in 2024 so you can try the web app in the meantime to get notified of its release. Soon you will be able to convert vocals in real-time in your DAW - they will even work for live performances!