Loading page
AudioLens's license and the terms behind the models and datasets it ships with.
AudioLens itself is MIT licensed. The models and datasets it works with are a mix of permissive and non-commercial licenses — read this page before you deploy AudioLens anywhere other than your own machine, and especially before deciding what a public instance can show.
AudioLens's code is MIT licensed.
AudioLens does not redistribute any dataset audio. Datasets are fetched
or imported by scripts you run yourself, from each dataset's official source
— see Loading datasets and CLI & scripts. The
repository does include small selection manifests (scripts/manifests/):
CSVs listing which clips are used, with their labels and annotations. Those
are derived from the source datasets, so they keep the source license
(attribution always; share-alike for RAVDESS and SAA; non-commercial for
RAVDESS, SAA and L2-ARCTIC). The MIT license on
AudioLens's code says nothing about the license terms of the audio,
transcripts, or model weights it helps you work with — those are listed
below.
| Asset | License | Key terms | Source |
|---|---|---|---|
openai/whisper-base | Apache 2.0 | Permissive, commercial use OK | huggingface.co/openai/whisper-base |
openai/whisper-large-v3 | Apache 2.0 (same Whisper license family) | Permissive, commercial use OK | huggingface.co/openai/whisper-large-v3 |
r-f/wav2vec-english-speech-emotion-recognition | Apache 2.0 | Permissive; fine-tuned from jonatasgrosman/wav2vec2-large-xlsr-53-english | huggingface.co/r-f/wav2vec-english-speech-emotion-recognition |
facebook/wav2vec2-base-960h (legacy fallback) | Apache 2.0 | Permissive | huggingface.co/facebook/wav2vec2-base-960h |
jonatasgrosman/wav2vec2-large-xlsr-53-english (legacy fallback) | Apache 2.0 | Permissive | huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-english |
| Common Voice (cv-valid-dev subset) | CC0 ("No Rights Reserved") — stated in the LICENSE.txt of the Kaggle dataset AudioLens uses, and by Mozilla | Public domain, commercial use OK. Mozilla's terms also ask you not to try to identify speakers. The subset is the Kaggle v1 release; a Kaggle account is needed to download it. Newer releases (23.0 onward) are distributed through the Mozilla Data Collective, whose live terms could not be fetched automatically, so check those separately if you use them | kaggle.com/datasets/mozillaorg/common-voice, commonvoice.mozilla.org |
| RAVDESS | CC BY-NC-SA 4.0 — non-commercial only | Commercial license must be purchased separately (contact ravdess@gmail.com); attribution + share-alike required; cite the PLoS ONE paper | zenodo.org/records/1188976 |
| L2-ARCTIC | CC BY-NC 4.0 — non-commercial only | Attribution required; must cite Zhao et al. (2018) Interspeech; registration (name/email/affiliation) required before download; commercial use needs approval from Dr. Ricardo Gutierrez-Osuna | psi.engr.tamu.edu/l2-arctic-corpus |
| Speech Accent Archive (SAA) | CC BY-NC-SA 4.0 (stated on the archive's own site) | Non-commercial, attribution, share-alike; cite Weinberger, S. H. (2015), Speech Accent Archive, George Mason University | accent.gmu.edu |
| LibriSpeech / LibriSpeech-1000 | CC BY 4.0 | Attribution required, commercial use allowed | openslr.org/12, huggingface.co/datasets/openslr/librispeech_asr |
| SAVEE | Unverified — check source | The project's own import script states plainly that "SAVEE is licensed and cannot be redistributed" — not freely licensed; requires obtaining a copy directly from the University of Surrey. Automated fetch of the official SAVEE site failed (TLS certificate mismatch) | kahlan.eps.surrey.ac.uk/savee (fetch blocked) |
Backend/data stay until you delete them; uploaded datasets and
sessions expire after 24 hours.See Loading datasets for how each dataset gets into AudioLens, and Hosted demo for how the demo uses these datasets as a non-commercial deployment.