How to make Hatsune Miku NT1.0 sing in somewhat fluent english

 BEFORE YOU START

This guide will only teach you how to correct pronunciation, there aren't really any concrete rules on pitch or whatever else tuning you'll be using this for. Hatsune Miku NT is a Japanese voice library that doesn't have any English capabilities whatsoever and this is a pretty involved process since you'll have to fully tune your main vocals in two separate engines, but I think the end result is definitely worth it.

If you're down with all that, let's get started.

VOWEL CONVERSION

There have been several guides before on how to make Japanese VOCALOIDs sing in english through phoneme editing, but through a very specific parameter exclusive to Piapro Studio NT parameter known as "attack speed" we can actually get a better result through the fact that it somewhat "blends" vowels together. (not to be confused with the way the Asp"glitch works in VOCALOID2) so I'd like to chart out the most common English phonemes and their closest Japanese equivalent.
 
The empty space in the phoneme conversion indicates which note should be longer to properly blend and emphasize the sound we want. It's important that for each conversion with more than one vowel you turn down the attack speed parameter all the way on all of them except for the first one.


Phonemes coloured in RED will need to be spliced with an outside source to sound fluent

[Phoneme(X-SAMPA/ARPABET)]  
[Conversion]
[V]/[ah] [M][a  ]
[e]/[eh] [e]
[I]/[ih] [i][e  ]
[i:]/[iy] [i ][e] 
[{]/[ae] [e ][a] 
[O:]/[ao] [o][a ] 
[Q]/[aa] [o][a ] 
[U]/[uh] [M ][e] 
[u:]/[uw] [M] 
[@r]/[er] [a ][M] 
[eI]/[ey] [e ][i] 
[aI]/[ay] [a ][e] 
[OI]/[oy] [o ][i] 
[@U]/[ow] [a][o ][M] 
[@r]/[er] [a ][M] 
[I@]/[iy r] [i ][a] 
[e@]/[eh r] [e ][M] 
[O@]/[ao r] [o][a ][M] 
[Q@]/[aa r] [o][a ][M] 

CONSONANT CONVERSION

Consonants are thankfully a little less involved but if you want to get more detailed with this section you can. I'll be providing a basic chart for phoneme conversions but none of these are concrete and can definitely change with context, but if I were to cover every possible consonant combination this page would be a mile long ^^.

I've left out phonemes that are the same across both phonetic systems and don't need to be edited or replaced.


Phonemes coloured in RED will need to be spliced with an outside source to sound fluent

[Phoneme(X-SAMPA/ARPABET)]  
[Conversion]
[r]/[r] [w]
[l0]/[l] [4]
[v]/[v] [m]
[T]/[th] [s]
[D]/[dh] [4]
[Z]/[zh] [S]
[t]/[t] [ts]
[d]/[d] [dz]

SPLICING

For this section you'll need to find a feminine vocalsynth or a real vocalist that's more fluent in English than Hatsune Miku is that also won't sound out of place if you splice parts of them into her. I'll be using RX Audio Editor for this section. 

For vowel and voiced consonant splicing, both your original .ppsf file and whatever file you use for this new vocal will need to be tuned exactly the same way, or none of this will blend together properly. What we're doing here is essentially an auditory illusion, we want to take just enough out of a secondary vocal to support pronunciation, but not so much that it distracts from the base tone of Hatsune Miku NT.

In RX Audio editor, open up your Hatsune Miku NT render, Normalize her to -1.0dB and go to the section that needs vowel splicing.

 



Next, open up your secondary vocalist that you're gonna splice from. Normalize this to -1.0dB. This can be literally any engine or vocalist as long as the tone is somewhat Miku-like. We're gonna try to preserve as much of the Piapro NT engine noise as we can here by only selecting a small portion of the vowel. (try to only take 3 or 4 lines from the bottom of the spectrogram or the illusion will be ruined)


 
 

For most of consonants, you can just take them directly from the secondary vocalist like this. But be wary of voiced consonants and try not to use too many of them.




Once you've done all this you should have an output that sounds pretty presentable. Finish adding any final touches and you're done!

Comments

Post a Comment

Popular posts from this blog

Basic Vocal Mixing tutorial

Image
For this basic tutorial I'll try to only use the most default plugins in a DAW so that this method can be accessible to everyone. If you want a version more advanced I will make an update later detailing my regular process!! Plugins and Their Order 1. EQ (for cutting frequencies) 2. Multiband Compressor (for compressing the low-mid frequencies only) 3. Compressor (for making the vocal stand out above the instrumental) 4. EQ (for boosting frequencies) 5. Limiter (for catching what the compressor couldn't) 6. a De-esser of some sort (for de-essing, duh!) 7. Reverb (for making the vocals sound less dry) it's very important to place them specifically in this order, as the plugins affect eachother as long as they're *ABOVE* one another (took me a while to realize this ^^;)   1. EQ This EQ Will be used for cutting low frequencies, try not to use more than 3 points/bands unless you absolutely have to! Remember, the tip to good EQing is that less is usually more.  The most cruc...
Read more

Step-by-step process of how I use Hatsune Miku NT

Image
 STEP 1. TUNE EVERYTHING This section is going to be unnecessarily detailed and probably slightly intimidating but bear with me. Piapro Studio NT has certain quirks and oddities that I make sure to iron out before I do anything else, so below I'll provide a list of pronunciation tips to make Hatsune Miku NT sound slightly less stilted, of course these aren't the end all be all of natural pronunciation and I am also NOT a native 日本語 speaker, but I hope these help regardless. CONSONANTS SUCH AS D, B AND G By default, Piapro Studio NT will try to space every consonant out to be relatively the same length, this can lead to some issues with sounds being awkwardly drawn out or the attack sounding odd.  For example: The consonants at the start of these notes are going to be unnecessarily loud and unnatural sounding, D and B specifically don't usually have a visible "pop" at the start of the waveform.  In order to fix this, we'll have to do the following: For "d...
Read more