The base model pretrained on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16Khz. Note: This model does not have a tokenizer as it was pretrained on audio alone.
SAVRN's Notes on wav2vec2-base
Sixteen kilohertz in, speech representations out. That is the contract with this base model from AI at Meta: pretrained on audio alone, shipped without a tokenizer, and a speech recognizer only once you build one and fine-tune on labeled text. Eight PyTorch files total 380 MB, so the hardware question is your fine-tuning run, not holding the weights. Our Index has no memory row or token price for it, which fits a model you fine-tune rather than call.
Apache 2.0 allows commercial use, modification and redistribution, so a fine-tuned derivative can ship in a product provided the license, copyright notices and any NOTICE file travel with it and you state what you changed. Check two things: the pretraining corpus is librispeech_asr, so weigh how far your own audio sits from it, and the weights date to December 2021 and have not moved, which makes the fine-tune entirely your work.
Model Card
By AI at Meta, published under apache-2.0, revision 0b5b8e868dd8.
The base model pretrained on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16Khz.
Note: This model does not have a tokenizer as it was pretrained on audio alone. In order to use this model speech recognition, a tokenizer should be created and the model should be fine-tuned on labeled text data. Check out this blog for more in-detail explanation of how to fine-tune the model.
Authors: Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli
Configuration
- Architecture
- Wav2Vec2ForPreTraining
- Layers
- 12
- Hidden size
- 768
- Feed-forward size
- 3,072
- Attention heads
- 12
- Vocabulary size
- 32
- Model type
- wav2vec2
Identity and Version
- Repository
- facebook/wav2vec2-base
- Publisher
- AI at Meta
- Task
- Not stated by the source
- Modality
- Other
- Library
- transformers
- Parameters
- Not stated by the source
- Languages
- en
- Revision
- 0b5b8e868dd84f03fd87d01f9c4ff0f080fecfe8
- First published
- 2022-03-02
- Last updated
- 2021-12-28
Files and Weights
8 files, 380.3 MB in total. The weights are 1 file totalling 380.3 MB in bin.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| pytorch_model.bin | Weights | 380.3 MB | 3249fe98bfc6 |
| config.json | Configuration | 1.8 KB | — |
| preprocessor_config.json | Configuration | 159 B | — |
| special_tokens_map.json | Configuration | 85 B | — |
| README.md | Documentation | 2.0 KB | — |
| .gitattributes | Repository | 736 B | — |
| tokenizer_config.json | Tokenizer | 163 B | — |
| vocab.json | Tokenizer | 291 B | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 380.3 MB
Released by AI at Meta through its official repository on Hugging Face. Read the license.
Built From
- Described by arXiv:2006.11477
- Trained on (disclosed) librispeech_asr
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 380.3 MB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About wav2vec2-base
Can I use wav2vec2-base commercially?
Yes. wav2vec2-base 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.