UTau is still the only way to get free singing voices that don't sound like a broken calculator
I've been using UTau for about twelve years now. The interface looks like it was designed in 2005 because it was designed in 2005. That's also why it still works on any computer you pull out of a closet. Most people don't know where to start when they hear about free voicebanks, so let me walk through what actually matters before you waste a weekend trying to make a vocal sound tolerable. The main repositories are utau.5ch.net and doujinsanka.com, plus various individual voicebank creators who host their own pages. When you download a voicebank, you're getting a folder with phoneme WAV files, a character definition file, and sometimes a pitch parameter file. Extract the folder into C:\UTau\voice. That's it. The program reads whatever folders it finds there. No installer. No account. No email registration with half of them. I spent about three weeks in 2018 trying to make sense of the character sheet system before realizing it was just a text file that mapped consonant-vowel combinations to audio samples. Once you understand that structure, picking voicebanks becomes a matter of checking which phonemes are included and which are interpolated. Interpolated sounds are synthetic guesswork. UTau fills gaps by averaging neighboring phonemes. It sounds worse than you'd think, especially on rapid passages.
Unlock The Power Of Vocal Synthesis A Guide To Free Utau Voicebanks Download And Create
The actual creation side involves inputting a melody and lyrics into the UTau editor, then adjusting pitch, vibrato, and breath parameters per syllable. The learning curve is steep because the program gives you almost no guidance. You'll spend your first session watching a tutorial and realizing you need to manually adjust at least forty pitch points for a thirty-second phrase. It's tedious. The results can sound convincing once you've done it enough times. One thing nobody mentions when they're selling UTau as beginner-friendly: the default breathing and noise settings will make every note sound like it's being sung underwater unless you turn them down. I found this out after rendering my first full song and hearing what sounded like a chorus of people inhaling through their teeth. The fix is to go into the voicebank's .def file or the character settings and set the noise level to near zero, then manually add breath sounds between phrases only. You want breath, not constant air leakage. There's also the question of DTM integration. UTau can export individual WAV files per syllable, which is useful if you want to arrange in a DAW, but it can also output directly to a stereo mix if you route your audio through ASIO or WASAPI. The direct export path saves you from file management headaches, but you lose the ability to adjust individual syllables after rendering. I usually render at the syllable level and import everything into Reaper because I need the flexibility to fix timing drift later.
Another practical issue: some voicebanks include pitch bend data baked into the WAV files while others don't. If you load a voicebank with baked pitch bends and then apply your own pitch curve in UTau, the two will conflict and sound terrible. Check the voicebank creator's notes or just test one syllable and listen for warbling. If it warbles, either remove the external pitch curve or switch to a different voicebank for that passage. The biggest limitation of UTau is probably the lack of real-time performance. Every edit requires rendering, waiting, and listening. A single minute of vocals can take twenty minutes to render depending on your processor and how many voicebanks are in your library. I stopped trying to work in real-time years ago and just accept that vocal production in UTau is a batch process. You write the notes, render, fix the problem areas, render again. It's slow but it's free, and no paid software at this point offers true phoneme-level control without a subscription model attached.
Get the Full Details
