🎵 Audio-Separator by Politrees 🎵

Select the Model

Output Format

The format of the output audio file.

Override model default segment size instead of using the model default value.

Override segment size

Segment Size

Larger consumes more resources, but may give better results.

32 4000

Overlap

Amount of overlap between prediction windows. Lower is better but slower.

2 10

Pitch shift

Shift audio pitch by a number of semitones while processing. may improve output for deep/high vocals.

-24 24

Batch Size

Larger consumes more RAM but may process slightly faster.

1 16

Normalization threshold

The threshold for audio normalization.

0.1 1

Amplification threshold

The threshold for audio amplification.

0 1

Input Audio

Stem 1

Stem 2

Select the Model

Output Format

The format of the output audio file.

Override model default segment size instead of using the model default value.

Override segment size

Segment Size

Larger consumes more resources, but may give better results.

32 4000

Overlap

Amount of overlap between prediction windows. Higher is better but slower.

2 50

Pitch shift

Shift audio pitch by a number of semitones while processing. may improve output for deep/high vocals.

-24 24

Batch Size

Larger consumes more RAM but may process slightly faster.

1 16

Normalization threshold

The threshold for audio normalization.

0.1 1

Amplification threshold

The threshold for audio amplification.

0 1

Input Audio

Stem 1

Stem 2

Select the Model

Output Format

The format of the output audio file.

Enable denoising after separation.

Denoise

Hop Length

Usually called stride in neural networks; only change if you know what you're doing.

32 2048

Segment Size

Larger consumes more resources, but may give better results.

32 4000

Overlap

Amount of overlap between prediction windows. Higher is better but slower.

0.001 0.999

Batch Size

Larger consumes more RAM but may process slightly faster.

1 16

Normalization threshold

The threshold for audio normalization.

0.1 1

Amplification threshold

The threshold for audio amplification.

0 1

Input Audio

Stem 1

Stem 2

Select the Model

Output Format

The format of the output audio file.

Identify leftover artifacts within vocal output; may improve separation for some songs.

Post Process

Enable Test-Time-Augmentation; slow but improves quality.

TTA

Mirror the missing frequency range of the output.

High End Process

Post Process Threshold

Threshold for post-processing.

0.1 0.3

Window Size

Balance quality and speed. 1024 = fast but lower, 320 = slower but better quality.

320 1024

Agression

Intensity of primary stem extraction.

1 100

Batch Size

Larger consumes more RAM but may process slightly faster.

1 16

Normalization threshold

The threshold for audio normalization.

0.1 1

Amplification threshold

The threshold for audio amplification.

0 1

Input Audio

Stem 1

Stem 2

Select the Model

Output Format

The format of the output audio file.

Enable segment-wise processing.

Segment-wise processing

Segment Size

Size of segments into which the audio is split. Higher = slower but better quality.

1 100

Overlap

Overlap between prediction windows. Higher = slower but better quality.

0.001 0.999

Shifts

Number of predictions with random shifts, higher = slower but better quality.

0 20

Normalization threshold

The threshold for audio normalization.

0.1 1

Amplification threshold

The threshold for audio amplification.

0 1

Input Audio

Stem 1

Stem 2

Stem 3

Stem 4

Stem 5

Stem 6

Directory to cache model files

The directory where model files are stored.

File output directory

The directory where output files will be saved.

Keys for automatic determination of input file names, stems, and models to simplify the construction of output file names.

Keys:

NAME - Input File Name
STEM - Stem Name (e.g., Vocals, Instrumental)
MODEL - Model Name (e.g., BS-Roformer-Viperx-1297)

Example:

Usage: NAME_(STEM)_MODEL

Output File Name: Music_(Vocals)_BS-Roformer-Viperx-1297

Vocals Stem

Output example: Music_(Vocals)_BS-Roformer-Viperx-1297

Instrumental Stem

Output example: Music_(Instrumental)_BS-Roformer-Viperx-1297

Other Stem

Output example: Music_(Other)_BS-Roformer-Viperx-1297

Drums Stem

Output example: Music_(Drums)_BS-Roformer-Viperx-1297

Bass Stem

Output example: Music_(Bass)_BS-Roformer-Viperx-1297

Guitar Stem

Output example: Music_(Guitar)_BS-Roformer-Viperx-1297

Piano Stem

Output example: Music_(Piano)_BS-Roformer-Viperx-1297

List filter

Filter and sort the model list by 'stem'

List limit

Limit the number of models shown.

1 10

This Space created by Politrees.

python-audio-separator by beveradb.
Thanks to NeoPy for the help with the code.

Built with Gradio logo