ESC-50
Models identify an environmental sound from a short recording.
The benchmark groups its recordings into 50 sound classes.
- Made by
- Karol J. Piczak
- Released
- October 2015
- Kind
- Audio understanding and voice
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Prediction
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Audio and speech
Natural Language Supervision for General-Purpose Audio Representations tops the leaderboard with 98.3%
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
- 98.3 Natural Language Supervision for General-Purpose Audio Representations
- 98.1 BEATs Audio Pre-Training with Acoustic Tokenizers
- 97.0 HTS-AT A Hierarchical Token-Semantic Audio Transformer for Sound Classification and Detection
- 96.7 CLAP Learning Audio Concepts From Natural Language Supervision
- 96.4 CAT Causal Audio Transformer for Audio Classification
- 95.7 AST Audio Spectrogram Transformer
- 95.7 Connecting the Dots between Audio and Text without Parallel Data through Visual Knowledge Transfer
- 94.1 A Sequential Self Teaching Approach for Improving Generalization in Sound Event Recognition
- 94.0 A mixture model-based real-time audio sources classification method
- 92.3 Efficient End-to-End Audio Embeddings Generation for Audio Classification on Target Applications
- 89.5 Urban Sound Tagging using Multi-Channel Audio Feature with Convolutional Neural Networks
- 88.7 An Ensemble of Convolutional Neural Networks for Audio Classification
- 87.1 Environmental Sound Classification on the Edge A Pipeline for Deep Acoustic Networks on Extremely Resource-Constrained Devices
- 86.5 Unsupervised Filterbank Learning Using Convolutional Restricted Boltzmann Machine for Environmental Sound Classification
- 85.7 AclNet efficient end-to-end audio classification CNN
- 85.0 On Open-Set Classification with L3-Net Embeddings for Machine Listening Applications
- 84.9 Learning from Between-class Examples for Deep Sound Recognition
- 84.2 Novel Phase Encoded Mel Filterbank Energies for Environmental Sound Classification
- 83.5 Knowledge Transfer from Weakly Labeled Audio using Convolutional Neural Network for Sound Events and Scenes
- 82.6 Deep Multimodal Clustering for Unsupervised Audiovisual Learning
- 82.0 Novel TEO-based Gammatone Features for Environmental Sound Classification
- 79.8 Objects that Sound
- 79.3 Look, Listen and Learn
- 79.1 Learning Environmental Sounds with Multi-scale Convolutional Neural Network
- 74.2 Soundnet Learning sound representations from unlabeled video
- 73.2 Classifying environmental sounds using image recognition networks
- 71.0 Learning environmental sounds with end-to-end convolutional neural network (EnvNet)
- 68.5 Very Deep Convolutional Neural Networks for Raw Waveforms
- 66.3 WSNet Learning Compact and Efficient Networks with Weight Sampling
- 64.5 bar_chart Environmental Sound Classification with Convolutional Neural Networks - CNN baseline
- 64.3 auDeep Unsupervised Learning of Representations from Audio with Deep Recurrent Neural Networks
- 56.4 Comparison of Time-Frequency Representations for Environmental Sound Classification using Convolutional Neural Networks
- 1 Natural Language Supervision for General-Purpose Audio Representations HTSAT-22 model pretrained by natural language supervision 98.3%
- 2 BEATs Audio Pre-Training with Acoustic Tokenizers Transformer model pretrained with acoustic tokenizers 98.1%
- 3 HTS-AT A Hierarchical Token-Semantic Audio Transformer for Sound Classification and Detection Transformer model with hierarchical structure and token-semantic modules 97%
- 4 CLAP Learning Audio Concepts From Natural Language Supervision CNN model pretrained by natural language supervision 96.7%
- 5 CAT Causal Audio Transformer for Audio Classification Transformer model with MFMR features and a causal module 96.4%
- 6 AST Audio Spectrogram Transformer Pure Attention Model Pretrained on AudioSet 95.7%
- 7 Connecting the Dots between Audio and Text without Parallel Data through Visual Knowledge Transfer A Transformer model pretrained w/ visual image supervision 95.7%
- 8 A Sequential Self Teaching Approach for Improving Generalization in Sound Event Recognition Multi-stage sequential learning with knowledge transfer from Audioset 94.1%
- 9 A mixture model-based real-time audio sources classification method Dictionary of sound models used for classification (accuracy is computed on segments instead of files) 94%
- 10 Efficient End-to-End Audio Embeddings Generation for Audio Classification on Target Applications CNN model pretrained on AudioSet 92.3%
- 11 Urban Sound Tagging using Multi-Channel Audio Feature with Convolutional Neural Networks Pretrained model with multi-channel features 89.5%
- 12 An Ensemble of Convolutional Neural Networks for Audio Classification CNN ensemble with data augmentation 88.7%
- 13 Environmental Sound Classification on the Edge A Pipeline for Deep Acoustic Networks on Extremely Resource-Constrained Devices CNN model (ACDNet) with potential compression 87.1%
- 14 Unsupervised Filterbank Learning Using Convolutional Restricted Boltzmann Machine for Environmental Sound Classification CNN with filterbanks learned using convolutional RBM + fusion with GTSC and mel energies 86.5%
- 15 AclNet efficient end-to-end audio classification CNN CNN with mixup and data augmentation 85.7%
- 16 On Open-Set Classification with L3-Net Embeddings for Machine Listening Applications x-vector network with openll3 embeddings 85%
- 17 Learning from Between-class Examples for Deep Sound Recognition EnvNet-v2 ([tokozume2017a](http//www.mi.t.u-tokyo.ac.jp/assets/publication/LEARNING_ENVIRONMENTAL_SOUNDS_WITH_END-TO-END_CONVOLUTIONAL_NEURAL_NETWORK-poster.pdf)) + data augmentation + Between-Class learning 84.9%
- 18 Novel Phase Encoded Mel Filterbank Energies for Environmental Sound Classification CNN working with phase encoded mel filterbank energies (PEFBEs), fusion with Mel energies 84.2%
- 19 Knowledge Transfer from Weakly Labeled Audio using Convolutional Neural Network for Sound Events and Scenes CNN pretrained on AudioSet 83.5%
- 20 Deep Multimodal Clustering for Unsupervised Audiovisual Learning CNN + unsupervised audio-visual learning 82.6%
- 21 Novel TEO-based Gammatone Features for Environmental Sound Classification Fusion of GTSC & TEO-GTSC with CNN 82%
- 22 Objects that Sound Look, Listen and Learn (L3) network ([arandjelovic2017a](https//arxiv.org/pdf/1705.08168.pdf)) with stride 2, larger batches and learning rate schedule 79.8%
- 23 Look, Listen and Learn 8-layer convolutional subnetwork pretrained on an audio-visual correspondence task 79.3%
- 24 Learning Environmental Sounds with Multi-scale Convolutional Neural Network Multi-scale convolutions with feature fusion (waveform + spectrogram) 79.1%
- 25 Soundnet Learning sound representations from unlabeled video 8-layer CNN (raw audio) with transfer learning from unlabeled videos 74.2%
- 26 Classifying environmental sounds using image recognition networks 16 kHz sampling rate, GoogLeNet on spectrograms (40 ms frame length) 73.2%
- 27 Learning environmental sounds with end-to-end convolutional neural network (EnvNet) Combination of spectrogram and raw waveform CNN 71%
- 28 Very Deep Convolutional Neural Networks for Raw Waveforms 18-layer CNN on raw waveforms 68.5%
- 29 WSNet Learning Compact and Efficient Networks with Weight Sampling SoundNet 8-layer CNN architecture with 100x model compression 66.3%
- 30 bar_chart Environmental Sound Classification with Convolutional Neural Networks - CNN baseline CNN with 2 convolutional and 2 fully-connected layers, mel-spectrograms as input, vertical filters in the first layer 64.5%
- 31 auDeep Unsupervised Learning of Representations from Audio with Deep Recurrent Neural Networks MLP classifier on features extracted with an RNN autoencoder 64.3%
- 32 Comparison of Time-Frequency Representations for Environmental Sound Classification using Convolutional Neural Networks 3-layer CNN with vertical filters on wideband mel-STFT (median accuracy) 56.4%
Source: Official ESC-50 leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Natural Language Supervision for General-Purpose Audio Representations | 98.3% | HTSAT-22 model pretrained by natural language supervision | Official leaderboard | |
| 2 | BEATs Audio Pre-Training with Acoustic Tokenizers | 98.1% | Transformer model pretrained with acoustic tokenizers | Official leaderboard | |
| 3 | HTS-AT A Hierarchical Token-Semantic Audio Transformer for Sound Classification and Detection | 97% | Transformer model with hierarchical structure and token-semantic modules | Official leaderboard | |
| 4 | CLAP Learning Audio Concepts From Natural Language Supervision | 96.7% | CNN model pretrained by natural language supervision | Official leaderboard | |
| 5 | CAT Causal Audio Transformer for Audio Classification | 96.4% | Transformer model with MFMR features and a causal module | Official leaderboard | |
| 6 | AST Audio Spectrogram Transformer | 95.7% | Pure Attention Model Pretrained on AudioSet | Official leaderboard | |
| 7 | Connecting the Dots between Audio and Text without Parallel Data through Visual Knowledge Transfer | 95.7% | A Transformer model pretrained w/ visual image supervision | Official leaderboard | |
| 8 | A Sequential Self Teaching Approach for Improving Generalization in Sound Event Recognition | 94.1% | Multi-stage sequential learning with knowledge transfer from Audioset | Official leaderboard | |
| 9 | A mixture model-based real-time audio sources classification method | 94% | Dictionary of sound models used for classification (accuracy is computed on segments instead of files) | Official leaderboard | |
| 10 | Efficient End-to-End Audio Embeddings Generation for Audio Classification on Target Applications | 92.3% | CNN model pretrained on AudioSet | Official leaderboard | |
| 11 | Urban Sound Tagging using Multi-Channel Audio Feature with Convolutional Neural Networks | 89.5% | Pretrained model with multi-channel features | Official leaderboard | |
| 12 | An Ensemble of Convolutional Neural Networks for Audio Classification | 88.7% | CNN ensemble with data augmentation | Official leaderboard | |
| 13 | Environmental Sound Classification on the Edge A Pipeline for Deep Acoustic Networks on Extremely Resource-Constrained Devices | 87.1% | CNN model (ACDNet) with potential compression | Official leaderboard | |
| 14 | Unsupervised Filterbank Learning Using Convolutional Restricted Boltzmann Machine for Environmental Sound Classification | 86.5% | CNN with filterbanks learned using convolutional RBM + fusion with GTSC and mel energies | Official leaderboard | |
| 15 | AclNet efficient end-to-end audio classification CNN | 85.7% | CNN with mixup and data augmentation | Official leaderboard | |
| 16 | On Open-Set Classification with L3-Net Embeddings for Machine Listening Applications | 85% | x-vector network with openll3 embeddings | Official leaderboard | |
| 17 | Learning from Between-class Examples for Deep Sound Recognition | 84.9% | EnvNet-v2 ([tokozume2017a](http//www.mi.t.u-tokyo.ac.jp/assets/publication/LEARNING_ENVIRONMENTAL_SOUNDS_WITH_END-TO-END_CONVOLUTIONAL_NEURAL_NETWORK-poster.pdf)) + data augmentation + Between-Class learning | Official leaderboard | |
| 18 | Novel Phase Encoded Mel Filterbank Energies for Environmental Sound Classification | 84.2% | CNN working with phase encoded mel filterbank energies (PEFBEs), fusion with Mel energies | Official leaderboard | |
| 19 | Knowledge Transfer from Weakly Labeled Audio using Convolutional Neural Network for Sound Events and Scenes | 83.5% | CNN pretrained on AudioSet | Official leaderboard | |
| 20 | Deep Multimodal Clustering for Unsupervised Audiovisual Learning | 82.6% | CNN + unsupervised audio-visual learning | Official leaderboard | |
| 21 | Novel TEO-based Gammatone Features for Environmental Sound Classification | 82% | Fusion of GTSC & TEO-GTSC with CNN | Official leaderboard | |
| 22 | Objects that Sound | 79.8% | Look, Listen and Learn (L3) network ([arandjelovic2017a](https//arxiv.org/pdf/1705.08168.pdf)) with stride 2, larger batches and learning rate schedule | Official leaderboard | |
| 23 | Look, Listen and Learn | 79.3% | 8-layer convolutional subnetwork pretrained on an audio-visual correspondence task | Official leaderboard | |
| 24 | Learning Environmental Sounds with Multi-scale Convolutional Neural Network | 79.1% | Multi-scale convolutions with feature fusion (waveform + spectrogram) | Official leaderboard | |
| 25 | Soundnet Learning sound representations from unlabeled video | 74.2% | 8-layer CNN (raw audio) with transfer learning from unlabeled videos | Official leaderboard | |
| 26 | Classifying environmental sounds using image recognition networks | 73.2% | 16 kHz sampling rate, GoogLeNet on spectrograms (40 ms frame length) | Official leaderboard | |
| 27 | Learning environmental sounds with end-to-end convolutional neural network (EnvNet) | 71% | Combination of spectrogram and raw waveform CNN | Official leaderboard | |
| 28 | Very Deep Convolutional Neural Networks for Raw Waveforms | 68.5% | 18-layer CNN on raw waveforms | Official leaderboard | |
| 29 | WSNet Learning Compact and Efficient Networks with Weight Sampling | 66.3% | SoundNet 8-layer CNN architecture with 100x model compression | Official leaderboard | |
| 30 | bar_chart Environmental Sound Classification with Convolutional Neural Networks - CNN baseline | 64.5% | CNN with 2 convolutional and 2 fully-connected layers, mel-spectrograms as input, vertical filters in the first layer | Official leaderboard | |
| 31 | auDeep Unsupervised Learning of Representations from Audio with Deep Recurrent Neural Networks | 64.3% | MLP classifier on features extracted with an RNN autoencoder | Official leaderboard | |
| 32 | Comparison of Time-Frequency Representations for Environmental Sound Classification using Convolutional Neural Networks | 56.4% | 3-layer CNN with vertical filters on wideband mel-STFT (median accuracy) | Official leaderboard |
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy.
Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7%
- MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy.
Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy.
Qwen 3.8 Max 90.8%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules.
Qwen3.5-Omni-Plus 93.1%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the official leaderboard (32 entries).
- Pass Models scored 32 entries on its leaderboard, none a model we track.
- Note Newest score No scores yet.
- Note Frontier models No models scored yet.
- Note Headroom No scores yet.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 32 entries at the last read, none a model we track.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.