11–20 of 73 posts
In case someone wondered what does the title have to do with logic, it’s just probably a (common) malapropism. premise (noun) a previous statement or proposition from which another is inferred or follows as a conclusion. premises (noun) a house or building, together with its land and outbuildings, occupied by a business or considered in an official context.
[1] https://ai.facebook.com/blog/wav2vec-20-learning-the-structu...
Is the transcription of Macron's speech completely off or am I not understanding what the two shown texts represent?
Seems off indeed. Plus, the readme says French is not supported yet. Not the best demo IMO.
Cool project, seems like your model have similar WER as mine (4th reference in readme). Do you plan to do any pre-training on the encoder part in the future? Maybe something like this[1] [1] https://ai.facebook.com/blog/wav2vec-20-learning-the-structu...
[0] https://github.com/theblackcat102/Online-Speech-Recognition
In case someone wondered what does the title have to do with logic, it’s just probably a (common) malapropism. premise (noun) a previous statement or proposition from which another is inferred or follows as a conclusion. premises (noun) a house or building, together with its land and outbuildings, occupied by a business or considered in an official context.
You didn't have any issue understanding the original title.
What datasets are used to train the models?
LibriSpeech, Tatoeba, Common Voice and scraped YouTube videos.
Wondering if you might have a better reco of the French President if you have a model per dialect of English?
Having worked on ViaVoice OSX back in the day, we had to have models for different varieties of English. The US model couldn’t understand my northern English (think GoT) accent. It’s why the product came out with a UK localisation. Wondering if you might have a better reco of the French President if you have a model per dialect of English?
Having worked on ViaVoice OSX back in the day, we had to have models for different varieties of English. The US model couldn’t understand my northern English (think GoT) accent. It’s why the product came out with a UK localisation. Wondering if you might have a better reco of the French President if you have a model per dialect of English?
Yes, probably. The data I trained on mostly reflects UK and US accents.
As a FYI, I was told “the money” was in specific “dictionaries” for medical professionals and so forth. Apparently, doctors liked to dictate straight into text. Might be worth trying that $$$€€€£££?
Earlier quoted context omitted.
LibriSpeech, Tatoeba, Common Voice and scraped YouTube videos.
Do you get good results when adding scraped youtube audio? My model performance on LibriSpeech dev drops a bit when adding youtube audio to the training dataset ( my guess is likely due to poor alignment from auto generated captions ).