Search public pages, research tools, and SAVRN solutions.
Open-weight model · Audio classification
by AI at Meta facebook/audiobox-aesthetics
This model has been pushed to the Hub using the PytorchModelHubMixin integration: Unified automatic quality assessment for speech, music, and sound. Paper arXiv / MetaAI. Blogpost ai.meta.com This repository requires Python 3.9 and Pytorch 2.2 or greater.
What it takes to serve audiobox-aesthetics (104M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 0.2 GB | 0.2 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.1 GB | 0.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.1 GB | 0.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.
When a pipeline produces or ingests audio at volume and someone needs a score for how each clip sounds, this 104M-parameter model from AI at Meta returns one across speech, music and sound in a single pass. It is small enough to disappear into a rack: 0.2 GB of weights and 0.2 GB of memory at 16-bit, so the cheapest Index option, one MI300X with 192 GB at $1.85 an hour, is priced for the generator beside it, not for this. The repository is 6 files and 831 MB, safetensors only, and the code wants Python 3.9 and PyTorch 2.2 or newer.
CC BY 4.0 permits commercial use and adaptation with two obligations: credit the creator and state what you changed. Read the paper it is described by, arXiv:2502.05139, because the page lists no architecture, library or reported evaluations. Last update was March 4, 2025, so treat it as fixed.
By AI at Meta, published under cc-by-4.0, revision 9b1dd8e5df9a.
This model has been pushed to the Hub using the PytorchModelHubMixin integration: - Code: https://github.com/facebookresearch/audiobox-aesthetics - Paper: https://huggingface.co/papers/2502.05139
--- README below copied from https://github.com/facebookresearch/audiobox-aesthetics
Unified automatic quality assessment for speech, music, and sound.
Install via pip
pip install audiobox_aesthetics
Install directly from source
This repository requires Python 3.9 and Pytorch 2.2 or greater. To install, you can clone this repo and run:
pip install -e .
Model | S3 | HuggingFace |---|---|---| All axes | checkpoint.pt | HF Repo
6 files, 831.4 MB in total. The weights are 2 files totalling 831.0 MB in pt, safetensors.
| File | Type | Size | SHA-256 |
|---|---|---|---|
| checkpoint.pt | Weights | 415.5 MB | a4931a7a01c3 |
| model.safetensors | Weights | 415.5 MB | a5a3c2412649 |
| config.json | Configuration | 492 B | — |
| README.md | Documentation | 5.9 KB | — |
| assets/aes_model.png | Other | 389.1 KB | bcc57fff5a71 |
| .gitattributes | Repository | 1.6 KB | — |
Released by AI at Meta through its official repository on Hugging Face. Read the license.
| Precision | Weights in memory |
|---|---|
| As published | 831.0 MB |
| 16-bit | 0.2 GB |
| 8-bit | 0.1 GB |
| 4-bit | 0.1 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
About 0.2 GB at 16-bit and 0.1 GB at 4-bit: the weights (104M parameters) plus a working margin. A long context needs more.
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Yes. audiobox-aesthetics is released under Creative Commons Attribution 4.0. CC BY 4.0 permits sharing and adapting the work, including commercially, provided the creator is credited and changes are indicated.
Fine-tuned facebook/wav2vec2-base for audio classification of single drum/percussion sounds into 10 classes. - clap, conga, crash, cymbal, hat, kick, ride, rim, snare, tom - Trained on short, single-hit drum sounds. Performance may drop on long mixes, multiple overlapping sounds, or very different recording conditions.
This model is a fine-tuned version of facebook/wav2vec2-base-960h for Speech Emotion Recognition (SER). It has been trained using a Frozen Feature Extractor strategy to preserve the model's acoustic understanding while adapting to emotion detection. This approach ensures stable performance and prevents "Catastrophic Forgetting," achieving nearly 80% accuracy on the validation set. Update: The "Calm" and "Neutral" classes have been merged to improve classification consistency, resulting in 7 distinct emotion classes. The model was trained on a combined dataset of ~12,000 audio files from: The model classifies audio into one of the following emotions: 1. Angry 2. Disgust 3. Fear 4. Happy 5.…
Wav2Vec2: Self-Supervised Learning for Speech Recognition: https://arxiv.org/pdf/2006.11477 male female Common-Voice-Gender-Detection is designed for: Speech Analytics – Assist in analyzing speaker demographics in call centers or customer service recordings. Conversational AI Personalization – Adjust tone or dialogue based on gender detection for more personalized voice assistants. Voice Dataset Curation – Automatically tag or filter voice datasets by speaker gender for better dataset management. Research Applications – Enable linguistic and acoustic research involving gender-specific speech patterns. Multimedia Content Tagging – Automate metadata generation for gender identification in…
This model is a fine-tuned version of mo-thecreator/wav2vec2-base-finetuned on the None dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 3e-05 - trainbatchsize: 8 - evalbatchsize: 8 - gradientaccumulationsteps: 4 - totaltrainbatchsize: 32 - lrschedulertype: linear - lrschedulerwarmupratio: 0.1 - numepochs: 5 - Transformers 4.39.3 - Pytorch 2.1.2 - Datasets 2.18.0 - Tokenizers 0.15.2 - mo-thecreator
This model is a fine-tuned version of motheecreator/Deepfake-audio-detection on the audiofolder dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 3e-05 - trainbatchsize: 32 - evalbatchsize: 32 - gradientaccumulationsteps: 4 - totaltrainbatchsize: 128 - lrschedulertype: cosine - lrschedulerwarmupratio: 0.1 - numepochs: 5 - Transformers 4.41.2 - Pytorch 2.1.2 - Datasets 2.19.2 - Tokenizers 0.19.1
The model expects a raw audio signal as input and outputs predictions for age in a range of approximately 0...1 (0...100 years) and gender expressing the probababilty for being child, female, or male. In addition, it also provides the pooled states of the last transformer layer. The model was created by fine-tuning Wav2Vec2-Large-Robust Timit and For this version of the model we only trained the first six transformer layers. An ONNX export of the model is available from Further details are given in the associated paper and tutorial.