SAVRN Model Hub · Comparisons
clap-htsat-fused vs wav2vec2-large-robust-12-ft-emotion-msp-dim
Clap-htsat-fused has 154M parameters and wav2vec2-large-robust-12-ft-emotion-msp-dim has 165M parameters; clap-htsat-fused is released under Apache License 2.0 and wav2vec2-large-robust-12-ft-emotion-msp-dim under Creative Commons Attribution-NonCommercial-ShareAlike 4.0; at 16-bit, clap-htsat-fused needs about 0.4 GB (1x MI300X from $1.85 an hour) and wav2vec2-large-robust-12-ft-emotion-msp-dim about 0.4 GB (1x MI300X from $1.85 an hour).
| Field | clap-htsat-fused laion/clap-htsat-fused | wav2vec2-large-robust-12-ft-emotion-msp-dim audeering/wav2vec2-large-robust-12-ft-emotion-msp-dim |
|---|---|---|
| Publisher | LAION eV | audEERING GmbH |
| Task | Audio classification | Audio classification |
| Modality | Audio | Audio |
| Parameters, as reported | 154M parameters | 165M parameters |
| Architecture | ClapModel | Wav2Vec2ForSpeechClassification |
| Library | transformers | transformers |
| Context length | 514 tokens | Not stated |
| Repository size | 1.2 GB | 1.3 GB |
| Artifact formats | safetensors, pytorch | safetensors, pytorch |
| License | apache-2.0 | cc-by-nc-sa-4.0 |
| Access | Open weights, no gate | Open weights, no gate |
| Memory at 16-bit (weights and margin) | 0.4 GB | 0.4 GB |
| Cheapest GPUs at 16-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Memory at 4-bit (weights and margin) | 0.1 GB | 0.1 GB |
| Cheapest GPUs at 4-bit, per hour | 1x MI300X, $1.85 | 1x MI300X, $1.85 |
| Revision viewed | 365dea6ef167 | 6eba34a2485e |
| Downloads reported by the hub | 8.2M | 804.8k |
| Last observed | 2026-09-18 | 2026-09-18 |
An evaluation row appears only where at least two of these models report the same benchmark with the same stated configuration, metric, unit and setup. Different evaluators stay named in each cell. Values are shown as reported: no unit conversion, no ranking.
SAVRN's Notes on clap-htsat-fused
Pairing sound with text is the job here: LAION eV's clap-htsat-fused puts audio and text embeddings in one space, so it can classify audio zero-shot against labels you write or produce features for search. Training data was LAION-audio-630k; size is 154M parameters. At 16-bit the weights are 0.3 GB and the run needs 0.4 GB; the cheapest listed option is one 192 GB MI300X at $1.85 an hour on-demand. Treat it as a sidecar on a card that is already serving something larger.
Apache 2.0 covers commercial deployment, modification and redistribution if the notices and any NOTICE file stay attached and significant changes are stated. Two checks before committing. The text side stops at 514 tokens, fine for labels and captions, not for long documents. And the repository is 1.23 GB in 11 files at float32 against 0.3 GB of 16-bit weights, so budget for storage and conversion.
SAVRN's Notes on wav2vec2-large-robust-12-ft-emotion-msp-dim
Three numbers come out of this model per clip: arousal, dominance and valence, each roughly 0 to 1. audEERING GmbH pruned the encoder from 24 to 12 layers before fine-tuning on MSP-Podcast, leaving 165M parameters. In 16-bit the weights are 0.3 GB and it runs in 0.4 GB, so the entry card, one MI300X at $1.85 per hour on-demand, is more than one instance needs; treat it as a rounding error on a busy card.
The license decides this one. CC BY-NC-SA 4.0 allows non-commercial sharing and adapting with credit, requires adaptations to carry the same license, and commercial use needs separate permission. The publisher calls the release research only and offers a commercial license for a model trained on much more data, so production use is a licensing conversation first. Check the paper, arXiv:2203.07378, and the ONNX export at doi:10.5281/zenodo.6221127 if you do not serve through transformers.
Questions
Which is larger, clap-htsat-fused or wav2vec2-large-robust-12-ft-emotion-msp-dim?
wav2vec2-large-robust-12-ft-emotion-msp-dim (165M parameters) is larger than clap-htsat-fused (154M parameters), by the parameter counts their publishers report.
Which is cheaper to run, clap-htsat-fused or wav2vec2-large-robust-12-ft-emotion-msp-dim?
At 4-bit, clap-htsat-fused fits on 1x MI300X from $1.85 an hour and wav2vec2-large-robust-12-ft-emotion-msp-dim on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use clap-htsat-fused commercially?
Yes. clap-htsat-fused is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
Can I use wav2vec2-large-robust-12-ft-emotion-msp-dim commercially?
Not without separate permission. wav2vec2-large-robust-12-ft-emotion-msp-dim is released under Creative Commons Attribution-NonCommercial-ShareAlike 4.0. CC BY-NC-SA 4.0 permits non-commercial sharing and adapting with credit, and requires adaptations to use the same license. Commercial use needs separate permission.