| #### Whisper Tiny (EN) | |
| - ID: openai/whisper-tiny.en | |
| - Hugging Face: [model](https://huggingface.co/openai/whisper-tiny.en) | |
| - Creator: openai | |
| - Finetuned: No | |
| - Model Size: 39 M Parameters | |
| - Model Paper: [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) | |
| - Training Data: The models are trained on 680,000 hours of audio and the corresponding transcripts collected from the internet. 65% of this data (or 438,000 hours) represents English-language audio and matched English transcripts, roughly 18% (or 126,000 hours) represents non-English audio and English transcripts, while the final 17% (or 117,000 hours) represents non-English audio and the corresponding transcript. This non-English data represents 98 different languages. | |
| @@ | |
| #### S2T Medium ASR | |
| - ID: facebook/s2t-medium-librispeech-asr | |
| - Hugging Face: [model](https://huggingface.co/facebook/s2t-medium-librispeech-asr) | |
| - Creator: facebook | |
| - Finetuned: No | |
| - Model Size: 71.2 M Parameters | |
| - Model Paper: [fairseq S2T: Fast Speech-to-Text Modeling with fairseq](https://arxiv.org/abs/2010.05171) | |
| - Training Data: [LibriSpeech ASR Corpus](https://www.openslr.org/12) | |
| @@ | |
| #### Wav2Vec Base 960h | |
| - ID: facebook/wav2vec2-base-960h | |
| - Hugging Face: [model](https://huggingface.co/facebook/wav2vec2-base-960h) | |
| - Creator: facebook | |
| - Finetuned: No | |
| - Model Size: 94.4 M Parameters | |
| - Model Paper: [Wav2vec 2.0: Learning the structure of speech from raw audio](https://ai.meta.com/blog/wav2vec-20-learning-the-structure-of-speech-from-raw-audio/) | |
| - Training Data: ? | |
| @@ | |
| #### Whisper Large v2 | |
| - ID: openai/whisper-large-v2 | |
| - Hugging Face: [model](https://huggingface.co/openai/whisper-large-v2) | |
| - Creator: openai | |
| - Finetuned: No | |
| - Model Size: 1.54 B Parameters | |
| - Model Paper: [Robust Speech Recognition via Large-Scale Weak Supervision](https://arxiv.org/abs/2212.04356) | |
| - Training Data: The models are trained on 680,000 hours of audio and the corresponding transcripts collected from the internet. 65% of this data (or 438,000 hours) represents English-language audio and matched English transcripts, roughly 18% (or 126,000 hours) represents non-English audio and English transcripts, while the final 17% (or 117,000 hours) represents non-English audio and the corresponding transcript. This non-English data represents 98 different languages. | |
| (evaluating this model might take a while due to it's size) |