RIP ELEVENLABS! Create BEST TTS AI Voices LOCALLY For FREE!

Summary

The video script introduces viewers to a comprehensive guide on creating custom text-to-speech (TTS) voices using AI on a local computer. The presenter, SK, outlines various methods ranging from a quick 10-second voice cloning technique to a more sophisticated, high-quality TTS model training process that requires only 2 minutes of audio. The video demonstrates how to install necessary software, use different web UIs for voice cloning and fine-tuning, and integrate the generated TTS audio with RVC (Reverse Voice Conversion) for enhanced voice quality. The ultimate goal is to enable users to produce high-fidelity TTS without incurring hefty fees for third-party services, offering a cost-effective solution for personalized voice generation.

Takeaways

• 🎉 You can create custom text-to-speech (TTS) AI voices on your local computer without paying high fees for pre-made AI voices.
• 🔧 There are various methods available, ranging from quick 10-second voice cloning to more sophisticated techniques for higher quality TTS.
• 📈 The process starts with installing necessary software like FFMpeg and Python, and can be done via one-click installers for patrons or manually.
• 📊 A graphic is provided to visualize the different methods for creating TTS voices, catering to different user needs and skill levels.
• ⏱ With just 10 seconds of audio, you can clone a voice using the XTTS web UI, which is the easiest and quickest method demonstrated.
• 📚 For better quality, you can train your own XTTS model using only 2 minutes of audio, which captures the nuances of the speaker's voice.
• 🔗 The training process is straightforward and does not require a powerful GPU, making it accessible for most users.
• 🤖 By using RVC (Reverse Voice Conversion), you can further improve the TTS audio to closely resemble the original voice, even of public figures.
• 🌐 There's an XTS-RVC UI that automates the process of generating TTS audio and then converting it with RVC, simplifying the workflow.
• 📝 The final 'Uber' method combines fine-tuned XTTS models with RVC for the highest quality and authenticity in TTS voice generation.
• 💾 Once you have your custom TTS model, you can use it without limitations, making it a cost-effective solution for voice generation needs.
• 📁 The script provides guidance on how to install and use the different tools, including tips for patrons and manual installation steps for all users.

Q & A

What is the purpose of the video?

The video aims to guide viewers on how to create custom text-to-speech AI voices on their local computer using various methods, from quick cloning with a short audio clip to training a more sophisticated model for higher quality results.

What are the two ways to install the required software for creating custom AI voices?

The two ways to install the required software are using the one-click installer available for Patreon supporters, which automatically installs FFMpeg and adds it to the path, and the manual way which requires having Python, FFMpeg, and the C++ build tools installed beforehand.

How long of an audio clip is needed for the simplest voice cloning method?

For the simplest voice cloning method, only 10 seconds of an audio clip is needed.

What is the minimum duration of audio required for training a custom text-to-speech model in the medium method?

In the medium method, a minimum of 2 minutes of audio is required for training a custom text-to-speech model.

How does the RVC software contribute to the final output of the text-to-speech process?

RVC (Resemblyzer Voice Converter) is used to further refine the generated text-to-speech audio by converting it to a voice that closely resembles a provided reference voice, significantly enhancing the quality and authenticity of the output.

What is the 'Uber text to speech method' and how does it differ from the medium method?

The 'Uber text to speech method' is a combination approach that involves using a fine-tuned XTTS model to generate audio and then importing that into RVC for further enhancement. It differs from the medium method by including the step of fine-tuning the model from scratch, which allows for more personalized and higher quality voice replication.

How can one obtain the PDF guide for remembering the steps to create custom AI voices?

The PDF guide can be obtained for free on the creator's Patreon page, which is linked in the video description.

What is the advantage of using the XTTS fine-tune web UI for training a custom model?

The XTTS fine-tune web UI allows users to train their own text-to-speech model using a relatively short audio clip. This enables the model to learn the specific accent, speaking style, and voice characteristics of the speaker, leading to a more personalized and accurate voice output.

What is the role of FFMpeg in the process of creating custom AI voices?

FFMpeg is a multimedia framework that is required for the installation process. It is used for handling various multimedia files and is essential for the proper functioning of the text-to-speech software.

How does the XTTS RVC UI automate the process of creating custom AI voices?

The XTTS RVC UI automates the process by integrating both the text-to-speech generation and the voice conversion steps into a single interface. Users can input text, select an RVC voice model, and upload a reference voice sample to automatically generate and convert the voice.

What is the significance of using a longer audio clip for training the text-to-speech model?

Using a longer audio clip, ideally around 10 minutes, provides the model with more data to learn from, which can result in a more accurate and higher quality replication of the speaker's voice.

How does the video help those who are tired of paying high fees for AI voice services?

The video provides a comprehensive guide on how to create custom AI voices without the need for expensive third-party services, empowering users to generate high-quality voice outputs at a fraction of the cost.

Outlines

00:00

😀 Introduction to Custom Text-to-Speech AI Voices

The video begins with an introduction to the process of creating custom text-to-speech AI voices. The host, SK, expresses enthusiasm about the various methods that will be covered, ranging from quick 10-second voice cloning to more advanced techniques for creating high-quality AI voices. The video offers a graphic to help viewers visualize the different methods that will be discussed. The host also provides instructions for installing necessary software, with options for both one-click installation for patrons and a manual installation process for others. The first method demonstrated is the quick cloning technique using just 10 seconds of audio.

05:02

🤖 Training Your Own Text-to-Speech Model

The second paragraph delves into training a custom text-to-speech model using only 2 minutes of audio. The host guides viewers through using the xtts fine-tune web UI, emphasizing that even a short audio clip can be used for training. The process involves creating a dataset, training the model with default settings, and optimizing the model for easier use. The host also mentions the importance of using the correct version of the software and provides a demonstration of the improved voice quality achieved through this method. The training allows for the replication of the speaker's accent, speech patterns, and other unique vocal characteristics.

10:04

🎤 Advanced Voice Cloning with RVC

The third paragraph introduces RVC, a powerful tool for voice cloning, which is used to further refine the text-to-speech output. The host explains that while RVC requires an initial audio file for conversion, it can produce highly accurate voice clones. The process involves generating text-to-speech audio and then using RVC to enhance it. The host also discusses a semi-automated method using the XTS RVC UI, which streamlines the process by automatically converting the text-to-speech audio with RVC. This method offers less manual control but is quicker and easier to use.

15:06

🚀 The Ultimate Text-to-Speech Method

The final paragraph describes the 'Uber text-to-speech' method, which combines the previous techniques to create an even higher quality voice output. The host demonstrates how to use a fine-tuned XTS model within the XTS web UI to generate audio, which is then further enhanced using RVC. The host also explains how to use the fine-tuned model with the XTS RVC UI for fully automated processing. The video concludes with a reminder that all the methods shown allow for the creation of high-quality, authentic-sounding AI voices without the need for expensive third-party software. The host encourages viewers to try out the methods and offers support through Patreon, where a PDF guide will also be made available.

Keywords

💡Text to Speech AI

Text to Speech AI refers to artificial intelligence systems that can convert written text into audible speech. In the video, it is the central theme as the host discusses various methods to create custom voices for this technology without incurring high fees.

💡Voice Cloning

Voice cloning is the process of replicating a person's unique voice using AI. The video demonstrates how to clone a voice with just 10 seconds of audio, which is a significant aspect of creating custom text to speech AI voices.

💡XTTS (eXtreme Text-to-Speech)

XTTS is a tool mentioned in the video used for text-to-speech conversion. It is used to create custom voice models and is a key component in the methods discussed for generating AI voices.

💡FFmpeg

FFmpeg is a free and open-source software project that handles multimedia data. In the context of the video, it is used as a part of the setup process for the text-to-speech systems discussed.

💡Python

Python is a high-level programming language that is widely used for various types of software development. In the video, it is mentioned as a prerequisite for manually installing the text-to-speech software.

💡Training a Model

Training a model in the context of the video refers to the process of teaching a machine learning algorithm to generate speech by providing it with sample audio data. This is a crucial step in creating a personalized text-to-speech voice.

💡RVC (Resemblyzer Voice Cloning)

RVC is a voice conversion software that can clone voices with high accuracy. It is used in the video to further refine the generated AI voice, making it sound more like the original speaker.

💡Web UI

Web UI stands for Web User Interface, which in the video refers to the graphical interfaces used to interact with the text-to-speech software without needing to use command-line instructions. They simplify the process of voice generation and model training.

💡Fine-tuning

Fine-tuning is a machine learning technique where a pre-trained model is further trained on a specific task to improve its performance. In the video, the host explains how to fine-tune an XTTS model using just 2 minutes of audio.

💡Batch File

A batch file is a script file in DOS, OS/2 and Windows that contains a series of commands to be executed by the command-line interpreter. In the video, batch files are used to automate the installation process of the text-to-speech software.

💡Deep Learning

Deep learning is a subset of machine learning that uses neural networks with many layers (hence 'deep') to analyze various factors of data. The video implies the use of deep learning in training the AI to accurately replicate human speech.

Highlights

Introduction to creating custom text-to-speech AI voices on your local computer without high fees.

Demonstration of various methods ranging from quick 10-second voice cloning to the ultimate text-to-speech voice.

Explanation on installing necessary software using one-click installer for Patreon supporters or manual installation.

Guidance on using FFMpeg and setting up the environment for text-to-speech voice generation.

How to clone a voice with just 10 seconds of audio using the XTTS web UI.

No character limit for the text input in the simple text-to-voice tab of the web UI.

Training your own text-to-speech model from scratch using only 2 minutes of audio.

Using Audacity to extend a short audio clip into a longer one for training purposes.

Details on the training process and the importance of using a longer audio clip for better results.

Optimizing the model for faster and more efficient use after training.

Replicating the accent, speech patterns, and unique quirks of a speaker in the generated audio.

Combining text-to-speech with RVC (Reverse Voice Conversion) for higher quality voice cloning.

Automatic conversion of text-to-speech audio using the XTS-RVC UI for ease of use.

Creating a custom Obama text-to-speech model and using it within the XTS web UI.

Achieving a highly authentic and quality text-to-speech output by fine-tuning and converting with RVC.

Availability of a PDF guide on Patreon for free to help remember the process.

Offer of priority support to Patreon supporters for any questions regarding the process.

Encouragement for viewers to try out the methods and have fun creating their own text-to-speech AI voices.

Transcripts

are you tired of the same old robotic

text to speech AI voices are you sick of

paying exorbitant fees for these AI

voices do you dream of creating your own

custom text to speech EA voices on your

own computer well today is your day

because this is the ultimate local text

speech AI video ever so that you can

replace yourself with the clickable

button hello humans my name is SK and oh

boy sit tight because today I'm going to

show you the ultimate way to get the

best text speech AI voices on your local

computer and I'll be showing you a range

of methods from the super lazy 10

seconds voice cloning to the ultimate

Uber absolute best text speech voice

possible I even made a little graphic to

help you visualize because of how many

things I'm going to show you today so

that no matter who you are and what your

goals are you will get the best results

possible to suit your needs so sit back

relax and let's begin okay so let's

start by installing all of these

softwares first and then I'll explain

all the different methods and to install

these you have two ways the first is of

course by using the one click installer

that is available for my Pat supporters

just download the two links of to your

computer then before the install you

need to actually launch the FFM Peg

install as admin this is the only file

that you need to run as administrator

because it will install FFM Peg and will

automatically add it to path and the

second is the ultimate text speech Auto

installer so just double click it and

then it will ask you which WB UI you

want to install now for this video I'm

going to choose four because I'm going

to install all of them but if you want

to install a particular wave UI for your

particular needs you can do it as well

and then you're going to press enter and

then will automatically begin the

installation oh and also after each

install is complete do not forget to

close the window press no if it asks you

so that the next installation can start

so that in the end all the three web UI

are installed automatically simple as

that you don't need to do anything and

the second way to install these is of

course the manual way so for this make

sure that you have python G for Windows

FFM Peg and the C++ build tools

installed onto your computer before

doing the installation so then first

we're going to inst inst the xtts withb

UI so you're going to click the

description down below you're going to

arrive on this page you're going to

click on this code button then on this

icon to copy this entire line then

inside the folder you're going to click

on the folder path type CMD press enter

and here we're going to clone the

repository so you're going to type get

clone and then paste the URL of the

repository then press enter then we're

going to go inside the folder and then

we're going to Simply largely install

the bat file and this will automatically

start the installation so after this is

done we're going to install xtts fine

tune webui which is very similar to the

first VII so once again click the second

link in the description down below then

click on code copy this entire URL then

once again clone the repository like we

did before then we go inside the folder

and then just like the other one we're

going to run the install.bat file so

then finally we're going to install the

third web UI so once again we're going

to copy the URL then copy the repository

then we going inside the folder then

we're going to create a new python

environment with this command then we're

going to activate the environment then

we're going to install torch and torch

audio with this command then we need to

install the requirements with this

command and there you go so now that we

have all of our wave UI installed where

do we actually begin well let's start

from the very beginning from the very

easiest that requires the least amount

of effort and work and that is the

simple quick cloning with 10 seconds of

audio that's right you only need 10

seconds of an audio clip to be able to

clone that voice and use it as much as

you want and we're going to do that

inside the XTS web UI so to do this

you're going to go inside the xtts withi

folder and then launch the start xtts

withi dobot file it will then download a

bunch of stuff and then the UI will

launch automatically so once you are

inside the wui you will see a bunch of

stuff but I can tell you right now

actually um this web UI is a little

broken meaning that the only real tab

that you can use is the simple text to

voice tab but don't worry it's

absolutely fine this is the only thing

that we really need anyway so then how

exactly do we use this well first it's

very simple all you need to do is just

input a text right here in put the text

that you want your voice to read so

something like I don't know hey guys

what's up or anything that you want

there is absolutely no character limit

but this is just like an example so then

you're going to scroll down you're going

to choose your language you do have a

choice of a bunch of different languages

which is really really cool then you're

going to scroll down here in this little

section you're going to upload your 10

seconds of voice clip where you can also

do less if you have like a voice clip

between 5 and 10 seconds this is more

than enough so like for example I

uploaded this clip from like an Obama

interview that is around like 38 seconds

long and in here without changing

anything you can just click and generate

and in only only a few seconds like it

took me like 2 seconds to generate we

get something feel like this hey guys

what's up and well there you go this is

it this is basically like the easiest

and laziest way to generate T to speech

from an existing audio clip and what's

really cool is that there's basically

like no limit of character I mean there

is probably a limit but it is very very

high so you can just like copy and paste

like this whole paragraph like the

script from the B Movie and then just

simply click generate and in only a few

seconds it should very quickly generate

your audio file ready to be used

according to all known laws of Aviation

there is no way a bee should be able to

fly so yeah there you go now obviously

this is not perfect but you will see

that very later on we will make it much

much better but as I say previously this

is the laziest way to create text to

speech on your local computer using the

quick cloning technique with 10 seconds

of audio now don't worry we're going to

come back to this with you a later but

for now this is the end of the first

method but how do we make it even better

well how about instead of using the

default text to speech xtts model we

train our own xtts model that's right

we're going to train our own text to

speech model from scratch which is why

this is called the medium text to speech

method we're going to find tune our xtts

model using only 2 minutes of audio

that's right not 10 minutes not 20

minutes only 2 minutes required and it's

actually really really easy and super

fast so for this we need to use the xtts

find tune web UI so you're going to go

inside the folder and then launch the

start.bat file and it will then give you

a local URL that you can just hold

control and then left click to open in

your browser so from this all you really

need as I said previously is only an

audio file with 2 minutes of audio and

that you're going to upload right here

now I'm actually going to give you a

little trick because although this is

only a 2minute voice file because of was

extremely lazy all I did is basically

took the 37 seconds of audio that I used

previously and all I did is put this

file inside audacity then I selected

this entire audio copied it and then

paste it multiple times one after the

other again and again and again until I

had a 2 minutes of audio that's right

that's all I really did so basically you

don't even need 2 minutes of continuous

audio to do a decent training now

obviously it is better if you do don't

be lazy like me if you really want a

good results you should definitely

strive for something like 10 minutes of

audio especially because you'll be able

to use this later on in the video but

once again if you're really really lazy

you can do it like me so basically once

you have imputed the audio you're going

to leave everything by default don't not

forget of course to choose your language

in my case it is English then you're

going to click on this little button

step one create data set so depending on

how long the audio file is the longer

the formatting will take and the

training doesn't even use a lot of vram

so pretty much anyone can use this so

you don't even need a very powerful GPU

to do this so that's really cool so

there you go for me it took less than 1

minute so now we can click on the second

tab where you basic basically going to

leave everything by default the only

thing that you can change if you really

want to is the number of epoch now six

for the number of epo is really like the

minimum one so you might want to

increase this something like 10 or maybe

12 but if this is the first time that

you train a XTS model I definitely

recommend leaving everything by default

because all of these are super optimized

already and they work really really well

also make sure that you always use the

2.0.2 version this is by far the best

version because the 2.0.3 is really not

as good then you going to click load

parameters from output folder so once

again just leave everything by default

and then simply click run the training

and then the training will start now I'm

not going to do it because I've already

done it before so I'm actually going to

you know stop the training but basically

once the training is finished it's just

going to say well training is done

you're then going to click on this

little optimize the model button that

will basically make the final files much

much smaller and easier to use and now

if we want to use this you're going to

click on the third tab called

interference click load parameters for

TTS from output folder it will say that

the parameters for TTS were loaded then

you going to click on this button to

load the model and then to test it out

basically just input the text and then

click interference and after a few

seconds we get something like this uh

this model sounds really good and above

all it's reasonably fast so yeah as you

heard this is much much better this is

very very close to the reference audio

that we used for training uh but

ultimately I had so much confidence now

the reason why we do this training is

that F Unity the XTS model allows you to

train on the accent of the speaker the

way the person speak the speed as well

as a few works so like for example if

you listen to the reference audio uh but

ultimately I had so much confidence in

as you can see there is a lot of like

poses like uh uh Etc and all of these

poses and quirks and sounds are

replicated inside the generated audio so

like for example listen to this hello uh

humans my name is k your AI Overlord uh

uh oh so yeah there you go as you heard

in the very beginning there is a lot of

these uh uh sounds that were present in

the reference audio that are now

replicated inside the generated audio so

doing this training doing this fine

tuning even with the small portion of

audio that we had it was enough for the

model to be able to replicate that voice

and now that you have this model you can

use it as much as you want there is

absolutely no limitation so yeah there

you go with have only 2 minutes of audio

which is actually only 30 seconds of

audio we still managed to train a pretty

cool text speed each model but how do we

make it even better that's right because

we're still not done if you remember

this was only the medium text to speech

method because now we can finally start

the ultimate text to speech combination

and inside that method there is even

three different methods that you can use

but essentially what we're doing is

taking the generated audio from text to

speech and putting it inside RVC to make

it even better now if you don't know

what RV see is I definitely recommend

you to watch this video first otherwise

you're not going to understand anything

from this point onward watching this

video is essential for the comprehension

of the rest of the video because RVC is

absolutely fantastic this is seriously

one of my favorite softwares of all time

because RVC allows you to basically

clone a voice to a basically perfect

level but RVC is only a voice to voice

conversion meaning that you need an

initial audio file before doing the

conversion and this is what we're going

to do inside method a which is a simple

conversion so inside the xtts web UI

you're going to input your text then put

the reference a your file and click

generate so that in the end we get

something like this hello everyone it's

me Barack Obama don't forget to

subscribe to my channel and leave a like

so once again not perfect but it is a

quick cloning so then you're going to

download the file so then you're going

to launch RVC once again if you are one

of my P supporters just use a one click

installer to install this and then from

this point on either you're going to

train your own voice from scratch as I

showed in my RVC video or download an

existing Voice already made by the

community so that you can then select it

inside the reference invoice then you're

going to copy the path of the file and

paste it right here and then simply

click convert and after a few seconds we

get something like this hello everyone

it's me Barack Obama don't forget to

subscribe to my channel and leave a like

so yeah as you heard now the voice is is

much much better I mean this is pretty

much the voice of Barack Obama and that

is probably because RVC is the most

powerful tool when it comes to cloning

voices so once you have created an audio

file you can easily convert it to any

voice that you want and you can of

course then use that file for anything

you want but all of that is a lot of

work I mean coming from xtts withi then

downloading the file putting it inside

RVC I mean it's nice and all but if only

there was a way to do it automatically

he hey boy well there actually is and

that is actually our next method using

the third web UI that I haven't talked

about yet called XTS RVC UI which

basically does everything that we just

did automatically so to use this you're

going to go inside the XTS RVC UI folder

then inside rvc's you're going to input

the RVC voice model which is basically

the pth file in the index file then

you're going to go back inside voices

you're going to input the reference

voice which is basically like the 10

seconds of audio that we use for cloning

and then you're going to back and launch

this sbot file and it will then give you

a local URL then you can hold control

then left click and then it will open to

web UI and then from here you're going

to choose the language the RVC model

then the voice sample then here you're

going to input your text you can leave

everything by default and then click

submit and in the end you should get

something like this hey guys what's up

it's me Barack don't forget to subscribe

to my channel and leave a like so yeah

basically as you see it first generated

an xtts audio and then automatically

converted with RVC so basically

everything that we did previously where

refer generated with xtts webui and then

put it inside RVC everything was done

automatically in one single click now it

is a nice Wii to use although compared

to the previous method it has a little

bit less functionality and parameters to

create the final audio that you want

however once again it is like a

compromise it requires less work so that

in the end you get some easy good

results and that's really pretty cool

but how do we make it even better that's

right we are still not done well let's

actually take a look at our guide and

let's take a look at our Uber text to

speech method and what is the Uber text

speech method well basically we're going

to take everything that we did

previously and put them all together so

basically we're going to take the fuud

xtts model to generate an audio inside

the xtts web UI and then import that

inside RVC because yes don't forget that

we fine tune a model from scratch and we

can use this whenever we want so for

this it's very simple we're going to go

inside the xtts fun webui folder go

inside F Chun models inside already and

this what you see right here is your

finetune model these files are your

finetune XTS models so what you're going

to do you're going to select them

control X to cut them then we're going

to go inside the xtts web UI inside

models now you're going to create a new

folder that you're going to name the

name of your speaker so in my case it is

Obama then we go inside and we paste

test our files right here and now if you

go back and we launch the webui now if

you click on select xtts model version

you will see our Obama XTS model that we

can now use inside the XTS web UI as

much as we want simple as that so once

again just input your text then input

the reference audio and now you can just

click generate and now we have our final

audio generated with our custom Obama

model inside the xtts web UI to give us

something like this hey guys so what do

you think of this pretty cool video am I

right subscribe right now come on which

is already very very close to the

original Obama voice but of course we

can make it even better by downloading

the file then launching RVC selecting

the reference voice model inputed the

path of the file just like last time and

now we can click convert and finally

after a few seconds we get our final

ultimate Uber text to speech audio file

listen to this hey guys so what do you

think of this pretty cool video on my

right subscribe right now come on I mean

come on this is really really good this

is pretty much the closest and most

powerful text to speech that you can do

on your local computer right now with

the highest level of quality the highest

level of authenticity and all of that

running on your local computer oh and

also since once again it is the same

model you can use your fin tun model

inside the XTS RVC UI so if you go ins

set models xtts copy those files and put

it like somewhere else in a separate

folder just in case and then if you use

like thein tune models once again just

copy those files inside the models xtts

folder contrl V replace all the files in

the destination and then it is ready to

be used automatically inside the xtts

RVC UI simple as that so yeah there you

go these were all the methods that you

can use to have the best text to speech

EA models running on your computer and

if you want to have like this little

graph so that you always remember what

to do I will make the PDF available for

free on my patreon in the description

down below and talk it off patreon do

not forget that I provide priority

support from my patreon supporters so if

you have any questions whatsoever do not

hesitate to send me a DM and I will try

to answer your question as soon as

possible the link for my patreon will be

in the description down below so yeah

there you go now you know pretty much

everything there is to know on how to

run the best T speech models running on

your local computer so that no matter

who you are and what are your project

you can do so without paying some

exorbitant fees for some third party

software so definitely try this out

yourself and have some fun and there we

are it folks thank you guys so much for

watching don't forget to subscribe and

smash the like button for the YouTube

algorithm thank you also so much to

supporters for supporting my videos you

guys are absolutely awesome you people

are literally the reason why I'm able to

make these videos so thank you so much

and I'll see you guys next time bye-bye