Applio

A simple, high-quality voice conversion tool focused on ease of use and performance.

Support — GitHub

Voice Model

Select the voice model to use for the conversion.

Index File

Select the index file to use for the conversion.

Upload Audio

Select Audio

Select the audio to convert.

Output Path

The path where the output audio will be saved, by default in assets/audios/output.wav

Export Format

Select the format to export the audio.

WAV MP3 FLAC OGG M4A

Speaker ID

Select the speaker ID to use for the conversion.

Split the audio into chunks for inference to obtain better results in some cases.

Split Audio

Apply a soft autotune to your inferences, recommended for singing conversions.

Autotune

Autotune Strength

Set the autotune strength - the more you increase it the more it will snap to the chromatic grid.

0 1

Clean your audio output using noise detection algorithms, recommended for speaking audios.

Clean Audio

Clean Strength

Set the clean-up level to the audio you want, the more you increase it the more it will clean up, but it is possible that the audio will be more compressed.

0 1

Enable formant shifting. Used for male to female and vice-versa convertions.

Formant Shifting

Post-process the audio to apply effects to the output.

Post-Process

Browse presets for formanting

Presets are located in /assets/formant_shift folder

Quefrency for formant shifting

Default value is 1.0

0 16

Timbre for formant shifting

Default value is 1.0

0 16

Apply reverb to the audio.

Reverb

Reverb Room Size

Set the room size of the reverb.

0 1

Reverb Damping

Set the damping of the reverb.

0 1

Reverb Wet Gain

Set the wet gain of the reverb.

0 1

Reverb Dry Gain

Set the dry gain of the reverb.

0 1

Reverb Width

Set the width of the reverb.

0 1

Reverb Freeze Mode

Set the freeze mode of the reverb.

0 1

Apply pitch shift to the audio.

Pitch Shift

Pitch Shift Semitones

Set the pitch shift semitones.

-12 12

Apply limiter to the audio.

Limiter

Limiter Threshold dB

Set the limiter threshold dB.

-60 0

Limiter Release Time

Set the limiter release time.

0.01 1

Apply gain to the audio.

Gain

Gain dB

Set the gain dB.

-60 60

Apply distortion to the audio.

Distortion

Distortion Gain

Set the distortion gain.

-60 60

Apply chorus to the audio.

chorus

Chorus Rate Hz

Set the chorus rate Hz.

0 100

chorus Depth

Set the chorus depth.

0 1

chorus Center Delay ms

Set the chorus center delay ms.

7 8

chorus Feedback

Set the chorus feedback.

0 1

Chorus Mix

Set the chorus mix.

0 1

Apply bitcrush to the audio.

Bitcrush

Bitcrush Bit Depth

Set the bitcrush bit depth.

1 32

Apply clipping to the audio.

Clipping

Clipping Threshold

Set the clipping threshold.

-60 0

Apply compressor to the audio.

Compressor

Compressor Threshold dB

Set the compressor threshold dB.

-60 0

Compressor Ratio

Set the compressor ratio.

1 20

Compressor Attack ms

Set the compressor attack ms.

0 100

Compressor Release ms

Set the compressor release ms.

0.01 100

Apply delay to the audio.

Delay

Delay Seconds

Set the delay seconds.

0 5

Delay Feedback

Set the delay feedback.

0 1

Delay Mix

Set the delay mix.

0 1

Select Custom Preset

Select file to import

Preset Name

Pitch

Set the pitch of the audio, the higher the value, the higher the pitch.

-24 24

Filter Radius

If the number is greater than or equal to three, employing median filtering on the collected tone results has the potential to decrease respiration.

0 7

Search Feature Ratio

Influence exerted by the index file; a higher value corresponds to greater influence. However, opting for lower values can help mitigate artifacts present in the audio.

0 1

Volume Envelope

Substitute or blend with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is employed.

0 1

Protect Voiceless Consonants

Safeguard distinct consonants and breathing sounds to prevent electro-acoustic tearing and other artifacts. Pulling the parameter to its maximum value of 0.5 offers comprehensive protection. However, reducing this value might decrease the extent of protection while potentially mitigating the indexing effect.

0 0.5

Hop Length

Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.

1 512

Pitch extraction algorithm

Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.

crepe crepe-tiny rmvpe fcpe hybrid[rmvpe+fcpe]

Embedder Model

Model used for learning speaker embedding.

contentvec chinese-hubert-base japanese-hubert-base korean-hubert-base custom

Select Custom Embedder

Folder Name

Upload .bin

Upload .json

The f0 curve represents the variations in the base frequency of a voice over time, showing how pitch rises and falls.

Please ensure compliance with the terms and conditions detailed in this document before proceeding with your inference.

I agree to the terms of use

Output Information

The output information will be displayed here.

Export Audio

Input Folder

Select the folder containing the audios to convert.

Output Folder

Select the folder where the output audios will be saved.

Export Format

Select the format to export the audio.

WAV MP3 FLAC OGG M4A

Speaker ID

Select the speaker ID to use for the conversion.

Split the audio into chunks for inference to obtain better results in some cases.

Split Audio

Apply a soft autotune to your inferences, recommended for singing conversions.

Autotune

Autotune Strength

Set the autotune strength - the more you increase it the more it will snap to the chromatic grid.

0 1

Clean your audio output using noise detection algorithms, recommended for speaking audios.

Clean Audio

Clean Strength

Set the clean-up level to the audio you want, the more you increase it the more it will clean up, but it is possible that the audio will be more compressed.

0 1

Enable formant shifting. Used for male to female and vice-versa convertions.

Formant Shifting

Post-process the audio to apply effects to the output.

Post-Process

Browse presets for formanting

Presets are located in /assets/formant_shift folder

Quefrency for formant shifting

Default value is 1.0

0 16

Timbre for formant shifting

Default value is 1.0

0 16

Apply reverb to the audio.

Reverb

Reverb Room Size

Set the room size of the reverb.

0 1

Reverb Damping

Set the damping of the reverb.

0 1

Reverb Wet Gain

Set the wet gain of the reverb.

0 1

Reverb Dry Gain

Set the dry gain of the reverb.

0 1

Reverb Width

Set the width of the reverb.

0 1

Reverb Freeze Mode

Set the freeze mode of the reverb.

0 1

Apply pitch shift to the audio.

Pitch Shift

Pitch Shift Semitones

Set the pitch shift semitones.

-12 12

Apply limiter to the audio.

Limiter

Limiter Threshold dB

Set the limiter threshold dB.

-60 0

Limiter Release Time

Set the limiter release time.

0.01 1

Apply gain to the audio.

Gain

Gain dB

Set the gain dB.

-60 60

Apply distortion to the audio.

Distortion

Distortion Gain

Set the distortion gain.

-60 60

Apply chorus to the audio.

chorus

Chorus Rate Hz

Set the chorus rate Hz.

0 100

chorus Depth

Set the chorus depth.

0 1

chorus Center Delay ms

Set the chorus center delay ms.

7 8

chorus Feedback

Set the chorus feedback.

0 1

Chorus Mix

Set the chorus mix.

0 1

Apply bitcrush to the audio.

Bitcrush

Bitcrush Bit Depth

Set the bitcrush bit depth.

1 32

Apply clipping to the audio.

Clipping

Clipping Threshold

Set the clipping threshold.

-60 0

Apply compressor to the audio.

Compressor

Compressor Threshold dB

Set the compressor threshold dB.

-60 0

Compressor Ratio

Set the compressor ratio.

1 20

Compressor Attack ms

Set the compressor attack ms.

0 100

Compressor Release ms

Set the compressor release ms.

0.01 100

Apply delay to the audio.

Delay

Delay Seconds

Set the delay seconds.

0 5

Delay Feedback

Set the delay feedback.

0 1

Delay Mix

Set the delay mix.

0 1

Select Custom Preset

Select file to import

Preset Name

Pitch

Set the pitch of the audio, the higher the value, the higher the pitch.

-24 24

Filter Radius

If the number is greater than or equal to three, employing median filtering on the collected tone results has the potential to decrease respiration.

0 7

Search Feature Ratio

Influence exerted by the index file; a higher value corresponds to greater influence. However, opting for lower values can help mitigate artifacts present in the audio.

0 1

Volume Envelope

Substitute or blend with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is employed.

0 1

Protect Voiceless Consonants

0 0.5

Hop Length

Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.

1 512

Pitch extraction algorithm

Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.

crepe crepe-tiny rmvpe fcpe hybrid[rmvpe+fcpe]

Embedder Model

Model used for learning speaker embedding.

contentvec chinese-hubert-base japanese-hubert-base korean-hubert-base custom

The f0 curve represents the variations in the base frequency of a voice over time, showing how pitch rises and falls.

Select Custom Embedder

Folder Name

Upload .bin

Upload .json

Please ensure compliance with the terms and conditions detailed in this document before proceeding with your inference.

I agree to the terms of use

Output Information

The output information will be displayed here.

Model Name

Name of the new model.

Architecture

Choose the model architecture:

RVC (V2): Default option, compatible with all clients.
Applio: Advanced quality with improved vocoders and higher sample rates, Applio-only.

RVC Applio

Sampling Rate

The sampling rate of the audio files.

32000 40000 48000

Vocoder

Choose the vocoder for audio synthesis:

HiFi-GAN: Default option, compatible with all clients.
MRF HiFi-GAN: Higher fidelity, Applio-only.
RefineGAN: Superior audio quality, Applio-only.

HiFi-GAN MRF HiFi-GAN RefineGAN

CPU Cores

The number of CPU cores to use in the extraction process. The default setting are your cpu cores, which is recommended for most cases.

1 16

GPU Number

Specify the number of GPUs you wish to utilize for extracting by entering them separated by hyphens (-).

GPU Information

The GPU information will be displayed here.

Dataset Path

Path to the dataset folder.

Dataset Creator

Dataset Name

Name of the new dataset.

Upload Audio Dataset

Audio cutting

Audio file slicing method: Select 'Skip' if the files are already pre-sliced, 'Simple' if excessive silence has already been removed from the files, or 'Automatic' for automatic silence detection and slicing around it.

Skip Simple Automatic

Chunk length (sec)

Length of the audio slice for 'Simple' method.

0.5 5

Overlap length (sec)

Length of the overlap between slices for 'Simple' method.

0 0.4

It's recommended to deactivate this option if your dataset has already been processed.

Process effects

It's recommended keep deactivate this option if your dataset has already been processed.

Noise Reduction

Noise Reduction Strength

Set the clean-up level to the audio you want, the more you increase it the more it will clean up, but it is possible that the audio will be more compressed.

0 1

Output Information

The output information will be displayed here.

Pitch extraction algorithm

Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.

crepe crepe-tiny rmvpe

Embedder Model

Model used for learning speaker embedding.

contentvec chinese-hubert-base japanese-hubert-base korean-hubert-base custom

Silent training files

Adding several silent files to the training set enables the model to handle pure silence in inferred audio files. Select 0 if your dataset is clean and already contains segments of pure silence.

0 10

Hop Length

Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.

1 512

Select Custom Embedder

Folder Name

Upload .bin

Upload .json

Output Information

The output information will be displayed here.

Batch Size

It's advisable to align it with the available VRAM of your GPU. A setting of 4 offers improved accuracy but slower processing, while 8 provides faster and standard results.

1 50

Save Every Epoch

Determine at how many epochs the model will saved at.

1 100

Total Epoch

Specifies the overall quantity of epochs for the model training process.

1 10000

Enabling this setting will result in the G and D files saving only their most recent versions, effectively conserving storage space.

Save Only Latest

This setting enables you to save the weights of the model at the conclusion of each epoch.

Save Every Weights

Utilize pretrained models when training your own. This approach reduces training duration and enhances overall quality.

Pretrained

Enable this setting only if you are training a new model from scratch or restarting the training. Deletes all previously generated weights and tensorboard logs.

Fresh Training

Cache the dataset in GPU memory to speed up the training process.

Cache Dataset in GPU

Enables memory-efficient training. This reduces VRAM usage at the cost of slower training speed. It is useful for GPUs with limited memory (e.g., <6GB VRAM) or when training with a batch size larger than what your GPU can normally accommodate.

Checkpointing

Utilizing custom pretrained models can lead to superior results, as selecting the most suitable pretrained models tailored to the specific use case can significantly enhance performance.

Custom Pretrained

Detect overtraining to prevent the model from learning the training data too well and losing the ability to generalize to new data.

Overtraining Detector

Upload Pretrained Model

Custom Pretrained G

Select the custom pretrained model for the generator.

Custom Pretrained D

Select the custom pretrained model for the discriminator.

Overtraining Threshold

Set the maximum number of epochs you want your model to stop training if no improvement is detected.

1 100

Index Algorithm

KMeans is a clustering algorithm that divides the dataset into K clusters. This setting is particularly useful for large datasets.

Auto Faiss KMeans

Please ensure compliance with the terms and conditions detailed in this document before proceeding with your training.

I agree to the terms of use

Output Information

The output information will be displayed here.

The button 'Upload' is only for google colab: Uploads the exported files to the ApplioExported folder in your Google Drive.

Exported Pth file

Pth file

Select the pth file to be exported

Exported Index File

Index File

Select the index file to be exported

Voice Model

Select the voice model to use for the conversion.

Index File

Select the index file to use for the conversion.

Applio is a Speech-to-Speech conversion software, utilizing EdgeTTS as middleware for running the Text-to-Speech (TTS) component. Read more about it here!

TTS Voices

Select the TTS voice to use for the conversion.

TTS Speed

Increase or decrease TTS speed.

-100 100

Text to Synthesize

Enter the text to synthesize.

Upload a .txt file

Input path for text file

Output Path for TTS Audio

Output Path for RVC Audio

Export Format

Select the format to export the audio.

WAV MP3 FLAC OGG M4A

Speaker ID

Select the speaker ID to use for the conversion.

Split the audio into chunks for inference to obtain better results in some cases.

Split Audio

Apply a soft autotune to your inferences, recommended for singing conversions.

Autotune

Autotune Strength

Set the autotune strength - the more you increase it the more it will snap to the chromatic grid.

0 1

Clean your audio output using noise detection algorithms, recommended for speaking audios.

Clean Audio

Clean Strength

Set the clean-up level to the audio you want, the more you increase it the more it will clean up, but it is possible that the audio will be more compressed.

0 1

Pitch

Set the pitch of the audio, the higher the value, the higher the pitch.

-24 24

Filter Radius

If the number is greater than or equal to three, employing median filtering on the collected tone results has the potential to decrease respiration.

0 7

Search Feature Ratio

Influence exerted by the index file; a higher value corresponds to greater influence. However, opting for lower values can help mitigate artifacts present in the audio.

0 1

Volume Envelope

Substitute or blend with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is employed.

0 1

Protect Voiceless Consonants

0 0.5

Hop Length

Denotes the duration it takes for the system to transition to a significant pitch change. Smaller hop lengths require more time for inference but tend to yield higher pitch accuracy.

1 512

Pitch extraction algorithm

Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.

crepe crepe-tiny rmvpe fcpe hybrid[rmvpe+fcpe]

Embedder Model

Model used for learning speaker embedding.

contentvec chinese-hubert-base japanese-hubert-base korean-hubert-base custom

Select Custom Embedder

Folder Name

Upload .bin

Upload .json

The f0 curve represents the variations in the base frequency of a voice over time, showing how pitch rises and falls.

Please ensure compliance with the terms and conditions detailed in this document before proceeding with your inference.

I agree to the terms of use

Output Information

The output information will be displayed here.

Export Audio

Voice Blender

Select two voice models, set your desired blend percentage, and blend them into an entirely new voice.

Model Name

Name of the new model.

Drag and drop your model here

Path to Model

You can also use a custom path.

Drag and drop your model here

Path to Model

You can also use a custom path.

Blend Ratio

Adjusting the position more towards one side or the other will make the model more similar to the first or second.

0 1

Output Information

The output information will be displayed here.

Download Model

Drag your plugin.zip to install it

Download Model

Model Link

Output Information

The output information will be displayed here.

Drop files

Drag your .pth file and .index file into this space. Drag one and then the other.

Download Pretrained Models

Pretrained

Select the pretrained model you want to download.

Sampling Rate

And select the sampling rate.

How to Report an Issue on GitHub

Click on the 'Record Screen' button below to start recording the issue you are experiencing.

Once you have finished recording the issue, click on the 'Stop Recording' button (the same button, but the label changes depending on whether you are actively recording or not).

Go to GitHub Issues and click on the 'New Issue' button.

Complete the provided issue template, ensuring to include details as needed, and utilize the assets section to upload the recorded file from the previous step.

Video

Path to Model

Introduce the model pth path

Output Information

The output information will be displayed here.

Upload Audio

Pitch extraction algorithm

Pitch extraction algorithm to use for the audio conversion. The default algorithm is rmvpe, which is recommended for most cases.

crepe fcpe rmvpe

F0 Curve

Image

Audio

Output Information

The output information will be displayed here.

Image

It will activate the possibility of displaying the current Applio activity in Discord.

Enable Applio integration with Discord presence

Theme

Select the theme you want to use. (Requires restarting Applio)

Version Checker

Check which version of Applio is the latest to see if you need to update.

Language

Select the language you want to use. (Requires restarting Applio)

Model Author Name

The name that will appear in the model information.

Output Information

The output information will be displayed here.

By using Applio, you agree to comply with ethical and legal standards, respect intellectual property and privacy rights, avoid harmful or prohibited uses, and accept full responsibility for any outcomes, while Applio disclaims liability and reserves the right to amend these terms.

Built with Gradio logo