Loading...
Describe a sound or provide reference audio to extract it from a recording and download it.
Input
Supports inputs in formats such as mp3 / wav / flac / ogg / m4a / aac / aif / aiff / mp4 / mov / mkv
Supported formats
Output
WAV
Use case
Remix, practice with accompaniment and edit separate parts.
Choose a model for your material and purpose, and review the examples before submitting.
Cost, time, formats and parameters depend on the selected model.
Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.
Specify a target in text using flow-matching separation; compare missed sounds and leakage with AudioSep on the same excerpt.
Describe the target sound in text and extract it from a mixture; useful when no reference audio is available.
CLAPSep: Remix, practice with accompaniment and edit separate parts.
Banquet: Remix, practice with accompaniment and edit separate parts.
SAM Audio Base: Remix, practice with accompaniment and edit separate parts.
SAM Audio Large: Remix, practice with accompaniment and edit separate parts.
ICML - 2026
Extract an open-vocabulary sound with AudioSep-Hive, FlowSep-Hive or CLAPSep, or isolate a source that matches a 10-second reference with Banquet.
IEEE/ACM TASLP - 2024
Extract an open-vocabulary sound with AudioSep-Hive, FlowSep-Hive or CLAPSep, or isolate a source that matches a 10-second reference with Banquet.
ISMIR - 2024
Extract an open-vocabulary sound with AudioSep-Hive, FlowSep-Hive or CLAPSep, or isolate a source that matches a 10-second reference with Banquet.
Includes public test results for models used here and elsewhere. The complete source table is shown by default; conditions and metrics can also be selected for comparison. Scores from different test conditions are not directly comparable.
Only records with verified public inference implementations and matching weights are shown; algorithms requiring no weights are also eligible. Closed services, unreleased weights and unmatched experiment versions are excluded. This is not an all-model leaderboard.
Public weights do not establish a fully open-source license. Expand the data for declared weight licenses and access conditions. Undeclared licenses are marked separately.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 3 / source rows 11
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
LASS-Net ( Liu et al., 2022 ) · Discriminative Models · CLIPSep ( Dong et al., 2023 ) · Discriminative Models · OmniSep ( Cheng et al., 2025 ) · Discriminative Models · ZETA ( Manor and Michaeli, 2024 ) · Generative Models · FlowSep ( Yuan et al., 2025 ) · Generative Models · ZeroSep ( Huang et al., 2025 ) · Generative Models · DGMO ( Lee et al., 2025b ) · Generative Models · SAM-Audio ( Shi et al., 2025 ) · Generative Models
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
The sources use different training or task settings, so compare numbers only within the stated protocol.
| Metric rank | Model | Test setting | Signal Fidelity ↑ | Perceptual & Semantic Quality | Reference-free Quality ↑ | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SDR | SI-SDR | LPAPS ↓ | FAD ↓ | CLAP-A ↑ | CLAP-T ↑ | MUSHRA ↑ | OQ | Rec | Pre | Fai | |||
| 1 | AudioSep (Hive)Public weights · Inference code | Discriminative Models | 5.67 | 5.02 | 3.86 | 0.80 | 0.65 | 0.31 | 68.4 | 3.34 | 4.81 | 3.40 | 4.39 |
| 2 | AudioSep ( Liu et al., 2024 )Public weights · Inference code | Discriminative Models | 2.37 | 1.58 | 4.22 | 0.90 | 0.57 | 0.26 | 60.9 | 2.94 | 4.86 | 2.96 | 4.47 |
| - | FlowSep (Hive)Used on this sitePublic weights · Inference code | Generative Models | - | - | 4.25 | 0.84 | 0.61 | 0.19 | 61.8 | 3.05 | 4.18 | 3.35 | 3.70 |
Markers identify models or components used here. Scores and settings come from the cited tests, not a new evaluation of the complete site pipeline.
Datasets, versions and conditions remain separate, including sources with no eligible ranking rows.
Deployable entries 3 / source rows 11
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
LASS-Net ( Liu et al., 2022 ) · Discriminative Models · CLIPSep ( Dong et al., 2023 ) · Discriminative Models · OmniSep ( Cheng et al., 2025 ) · Discriminative Models · ZETA ( Manor and Michaeli, 2024 ) · Generative Models · FlowSep ( Yuan et al., 2025 ) · Generative Models · ZeroSep ( Huang et al., 2025 ) · Generative Models · DGMO ( Lee et al., 2025b ) · Generative Models · SAM-Audio ( Shi et al., 2025 ) · Generative Models
Deployable entries 3 / source rows 3
Scores apply to the listed dataset, metric and model version; results from different tests are not directly comparable.
The table includes only models with public weights or configurations that can run locally.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 3 / source rows 11
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
LASS-Net ( Liu et al., 2022 ) · Discriminative Models · CLIPSep ( Dong et al., 2023 ) · Discriminative Models · OmniSep ( Cheng et al., 2025 ) · Discriminative Models · ZETA ( Manor and Michaeli, 2024 ) · Generative Models · FlowSep ( Yuan et al., 2025 ) · Generative Models · ZeroSep ( Huang et al., 2025 ) · Generative Models · DGMO ( Lee et al., 2025b ) · Generative Models · SAM-Audio ( Shi et al., 2025 ) · Generative Models
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | Test setting | Signal Fidelity ↑ | Perceptual & Semantic Quality | Reference-free Quality ↑ | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SDR | SI-SDR | LPAPS ↓ | FAD ↓ | CLAP-A ↑ | CLAP-T ↑ | MUSHRA ↑ | OQ | Rec | Pre | Fai | |||
| 1 | AudioSep (Hive)Public weights · Inference code | Discriminative Models | 5.67 | 5.02 | 3.86 | 0.80 | 0.65 | 0.31 | 68.4 | 3.34 | 4.81 | 3.40 | 4.39 |
| 2 | AudioSep ( Liu et al., 2024 )Public weights · Inference code | Discriminative Models | 2.37 | 1.58 | 4.22 | 0.90 | 0.57 | 0.26 | 60.9 | 2.94 | 4.86 | 2.96 | 4.47 |
| - | FlowSep (Hive)Public weights · Inference code | Generative Models | - | - | 4.25 | 0.84 | 0.61 | 0.19 | 61.8 | 3.05 | 4.18 | 3.35 | 3.70 |
The table includes only models with public weights or configurations that can run locally.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 3 / source rows 3
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | 2-sources separation · 1-target | 3-sources separation · 1-target | 3-sources separation · 2-target | 4-sources separation · 1-target | 4-sources separation · 2-taget | 4-sources separation · 3-target | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SDRi↑ | SISDRi↑ | SDRi↑ | SISDRi↑ | SDRi↑ | SISDRi↑ | SDRi↑ | SISDRi↑ | SDRi↑ | SISDRi↑ | SDRi↑ | SISDRi↑ | ||
| 1 | CLAPSep-PNPublic weights · Inference code | 12.67±7.80 | 11.79±8.23 | 8.96±6.13 | 8.09±6.17 | 8.90±5.78 | 7.82±6.32 | 7.73±5.79 | 6.80±6.00 | 5.20±4.63 | 4.01±4.95 | 8.00±4.87 | 6.73±5.20 |
| 2 | CLAPSep-PPublic weights · Inference code | 10.03±11.27 | 8.70±12.64 | 6.48±7.82 | 5.73±8.50 | 6.35±7.36 | 4.88±8.10 | 5.94±7.58 | 4.85±8.29 | 2.91±5.54 | 1.37±5.98 | 5.34±6.70 | 3.79±7.09 |
| 3 | AudioSepPublic weights · Inference code | 8.60±8.86 | 7.38±9.48 | 4.90±6.13 | 4.20±6.42 | 5.30±5.38 | 2.96±5.54 | 3.52±5.21 | 2.81±5.76 | 2.37±4.26 | 0.65±4.32 | 4.20±4.51 | 2.03±4.27 |