DeepSpeech | Speech to Text | Common Voice | Donate Your Voice 

That's a great video can you do a video about implementing it into your project

So do I understand this correctly: there is no actual application for Windows and Linus which uses deepspeech and is ready to translate my speech from mic into text while writing a text file? All of this is still in development?

Hello can i train deep speech on my national language? And any language in the world can be used as dataset? thank you

So basically... use it for short commands? Or does it get better. Sorry i am a lazy man who didnt watch the whole video

I think I would do spanish as my L1, and English as is my L2, but I am not sure if reading with a foreign accent will be any good contribution. ED** seems like is one of the main points xD

where can i get voices like the one in this video?

btw does anyone knows if there is a problem opening the common voice platform from the android version of firefox? cause the page doesn't load from a mobile device.

is there a way to install on mac? I need it to use with this software autoedit.gitbook.io/autoedit-3-user-manual/speech-to-text

how to deal with audio file above 16khz??

is deepspeech compatible with nvidia Rtx 3090?

This is promising but there are a few glaring defects in this approach. First I think the voice recognition is fairly good, but its pattern matching is horrible as it does not even attempt to match output to an exsisting real life word. To me this is the most glaring problem. There are very rare instances where you might want a word not in the English dictionary but these are very rare and could be switched off. Words like proofsless, almuch, aveenes, vordwardine, watol etc. should not being returned. Although it shows the user what is actually being heard, it is counterproductive and might influence the narrator to alter his or her speech to accommodate, thus defeating the purpose. This model is obviously dumbed down on purpose and gives a misleading accuracy rating by not pattern matching its words. Still something not quite right still requiring input when deepspeech has been available so long and computers can feed deepspeech voice data at a much higher rate than humans. Again I could see if they were asking for contributions for transcription, but not voice data, there is plenty of that around. I see voice data as relevant only if it is for your own personal recognition. Secondly the privacy issue of uploading your voice is really an intrusion into one's privacy and should be a choice. There already exists millions of hours of audio sources from tv, movies, interviews, commentary out there. It is quite suspect to want entrely new data. Asking people to transcribe already available sources would be much less intrusive on your privacy and provide a wider range of ambiant environments that this voice was recorded in. as an example: Sean Connery still has his same voice give or take but his voice has a whole different acoustic in a movie than it does a time he may have given a speech in an auditorium. Although individual contributions are more consistant, this is also its failing, especially since no audio standard has been demanded of contributors (based on some of the horrible noise in some of those recordings).

I want to say that and I don't mean it in a condescending way: you need to go out more and do sports outdoors.