How to use NEUTRINO (Offline & Online version)

How to use NEUTRINO (Offline & Online version)

The processing speed of the engine for Windows
CPU has been improved.
The processing speed of the GPU version of the engine has been improved.
Fixed the progress display of the CPU version.
Changed the default number of threads in Run.bat to 4 (NumThreads = 4)

macOS
Intel The processing speed of the engine of the CPU version of macOS has been improved.

The processing speed of the engine of the Online
CPU version has been improved.
The processing speed of the GPU version of the engine has been improved.
Changed the default number of threads in Run.sh to use all cores (NumThreads = 0)

**some CPUs will no longer operational.
(Pentium / Celeron series before Intel 11th generation and Core series before Intel 3rd generation)

**some GPUs will no longer operational.
(Some GTX700 series and GeForce series before GTX600 series)


Operating environment (Electron v1.0.0 ~)

Due to engine updates and speedups, some CPUs / GPUs are no longer in operation.
CPU: Pentium / Celeron series before Intel 11th generation and Core series GPU before Intel 3rd generation : Online version for CPUs / GPUs for some GTX700 series and GeForce series before GTX600 series, You can still using the old version (v.0.551).

About NSF (Electron v1.0.0 ~)
With Version.0.540, NSF is a new version and can be used in all environments. In addition, the expressiveness has improved and stable audio can be output from any singing voice library.

On the other hand, in the CPU version, optimization and parallel processing have not progressed yet, and it is a very heavy process. Consider turning off the NSF composition process if necessary.
Windows version: Comment out the NSF line of Run.bat (prefix it with ":").
macOS version: Comment out the NSF line in Run.sh (prefix it with "#").
Online version: Comment out the NSF line in Run.sh (prefix it with "#").

Memory usage depends on the length of the phrase in the score. The section enclosed by rests or breath symbols is one phrase. Please note that it may not work if the memory / GPU memory is low.

About the online version
In the online version, you can use all the functions of NEUTRINO. It works perfectly on your web browser and you can use high-speed rendering with Cloud GPU. It also works on smartphones, so please try it if you like.

Lyrics that can be entered (v.0.510 ~)
If the error message "There are undefined lyrics or notes with no lyrics entered. Please correct the lyrics of the note to hiragana / katakana." Is displayed, the sound generation itself will be stopped. ..
The bar number and lyrics of the corresponding note are displayed in the execution log of "musicXMLtoLabel", so please correct it to hiragana or katakana lyrics.
If nothing is displayed like "", there is a note with blank lyrics (no lyrics have been entered), so please enter the lyrics as appropriate.
For the characters that can be used, see the Hiragana / Phoneme list .

New option of WORLD (β version: v.0.510 ~)
Options for Pitch Smoothing (SmoothPitch), Formant Smoothing (SmoothFormant), and Enhance Breathiness. You can smooth the pitch and formants and strengthen the breath component, and you can set the strength of the effect from 0 to 100 [%] respectively.
While various changes can be made to the voice, the attack may be weakened or the noise may be increased. How to hang it depends on the singing voice library, so please adjust accordingly.

Style shift option (Advance)
It is an option for advanced users who perform a technique called "key change + pitch change gacha". Raise the entire score by N keys and type it in, and use the pitch change function to lower it by N keys.

NEUTRINO Please set "StyleShift" of Run.bat to a value (integer) of about -5 to 5 and try it. For example, if you set it to "2", it will sing the key two keys up, and if you set it to "-2", it will sing the key two keys down.

You can sing in the range you are good at, or you can sing in a calm manner with low keys.
It is also possible to output three patterns such as "StyleShift = -1", "StyleShift = 0", "StyleShift = 1" and pick up the good parts of each to make it look like multiple Takes were recorded. increase.

About GPU
[NVIDIA GPU (4GB or more GPU memory is recommended)] is required to use some functions of the Windows version (high-speed rendering by GPU). Please update the NVIDIA driver to the latest version before using.

Rendering with multiple GPUs is also supported, and the basic setting is the multi-GPU setting. By changing the option (-m) of NEUTRINO to a single GPU (-g), one GPU with the highest processing capacity is automatically selected. If the processing with multiple GPUs does not work, please try this.

If the NVIDIA GPU is not found, it will automatically move to CPU processing with a message such as "Run in CPU mode". You can operate in CPU mode by removing the option (-m) and adding the option (-c).

As of 2022 / 5, the required operating environment is the same for all singing voice libraries.
When considering the introduction of a limited distribution library (Akane Kotonoha, Aoi, etc.), it is sure to check the operation once with another singing voice library.

About GPU support for macOS
GPU support for macOS is currently pending.
It may be a transitional period to new generation CPUs and GPUs such as Apple Silicon
NEUTRINO ver.0.5 series update (Translate from japanese)

Version.0.5 series overview
This version will mainly improve quality and functionality toward the v.1.0 release.
v.0.100 ~ 0.320 was positioned as an α test version that implements basic functions in sequence, and v.0.400 ~ v.0431 was positioned as a β test version that increases the number of singing voice libraries, but v.0.5 series is v. It will be a version that will be improved and changed for 1.0 and GUI release.

[NVIDIA GPU (3GB or more GPU memory is recommended)] is required to use some functions of the Windows version (high-speed rendering by GPU, composition by NSF).
Please be assured that you can use it as an ordinary GPU for PC games and graphics without any problems.

You can use all the features of NEUTRINO in the online version. Since the operation is completed on the web browser, it does not require an NVIDIA GPU and can be operated from a smartphone. Read in resources for the online version setup procedure .

Version.0.500β update contents (2021/10/01)
  • The model of the singing voice library has been changed to a new method.
  • We have updated various models. (All singing voice library)
  • Various noises have been reduced.
  • The version of NEUTRINO and the version of the singing voice library are displayed at the time of execution.
  • * Due to the compatibility with models prior to Version.0.431 has been lost. When using a model earlier than v0.431, please use the one with the main unit version v.0.431.
_____________________________________________________________

Note
  • About NEUTRINO Voice Support Tool produced by sig, is a tool that allows you to edit timing, pitch, etc., but in the default state, you can only use "KIRITAN", "YOKO", and "KIRITAN_FAST". But you can use other model by renaming the model folder inside NEUTRINO Voice Support Tool file, you can also use the adjustment support tool with singing voice libraries such as Merrow and Tohoku Itako.
    Specifically, it works by changing "ITAKO" in the model folder to "KIRITAN" or "YOKO" or "KIRITAN_FAST" and selecting "KIRITAN" or "YOKO" or "KIRITAN_FAST" on the adjustment support tool side. increase. Please try it.
  • About style shift option (v.0.320 ~)
    This is an option for advanced users who perform a technique called "key change + pitch change". The entire score is raised by ** keys and typed in, and the pitch change function is used to lower the score by ** keys.
    NEUTRINO option called "StyleShift" has been added to Run.bat, so please set it to a value (integer) of about -5 to 5 and try it. For example, if you set it to "2", it will sing the key two keys up, and if you set it to "-2", it will sing the key two keys down.
    You can use in the range you are good at, or you can use in a calm manner with low keys. It is also possible to output three patterns such as "StyleShift = -1", "StyleShift = 0", "StyleShift = 1" and pick up the good parts of each to make it look like multiple takes were recorded. Please try.
  • About NEUTRINO (GPU) (v.0.300 ~)
    Rendering on multiple GPUs is also supported now. The basic setting is a multi-GPU setting. By changing the option (-m) of NEUTRINO to a single GPU (-g), one GPU with the highest processing capacity is automatically selected. If the processing on multiple GPUs does not work, please try this.
    If the NVIDIA GPU is not found, it will automatically move to CPU processing with a message such as "Run in CPU mode". If you do not want to see the message, remove the option (-m). You can force it to operate in CPU mode by removing the option (-m) and adding the option (-c). Please try if you encounter any problems related to GPU.
  • About NSF (v.0.200 ~)
    If you do not use NSF, comment out the NSF line of Run.bat (prefix it with ":"). If you don't have an NVIDIA GPU, comment out the NSF line as well.
    If multiple GPUs are stuck, you can select the GPU by changing "set CURRENNT_CUDA_DEVICE = 0" in NSF / Run.bat. Set according to the GPU number you are using. GPU memory usage depends on the length of the phrase in the score. The section surrounded by rests or breath symbols is one phrase, and it uses about 100MB per second. For example, if there is a phrase of 15 seconds, it will be calculated to use about 1.5 GB, and if the GPU memory is 2 GB, it may not work, so please be careful.
    In Version.0.500, the NSF version is not very compatible with the singing voice library (Merou, Nakumo, Tohoku Zunko, JSUT) that has a lot of breath components. We recommend using the WORLD version of the above library.

  • About GPU support for macOS
    GPU support and NSF porting for macOS is currently pending. Since it is a transitional period to new generation CPUs and GPUs such as Mac M1 chips, we have been watching for a while. You can run all the functions in the online version, so please use that if you like.
- Minor grammar change & word correction
- Add link to Neutrino Editor (Download & How to use)

NEUTRINO Online version guide is complete.

I hope that you all will experience the effects of deep learning through singing voice synthesis, and that it will be an opportunity for as many people as possible to start deep learning and machine learning.

How to use NEUTRINO

Have fun and Stay productive!

1. Correcting grammar error
2. re-write. make it more easy to understand

If you find any kind of issue, Please let me know.
NEUTRINO offline version guide is complete.

Since this is a lot of text and a lot of translation. There is a chance that i forgot something important or i write it incorrectly.

I will do online version asap when i have sparetime.

If you find this kind of issue, Please let me know.