31–40 of 73 posts
Earlier quoted context omitted.
Data and compute are the largest hurdles. I only have one GPU and training one model takes 3+ days, so I am limited by that. Also, scraping from YouTube takes time and a lot of storage (multiple TBs). Mozilla Common Voice data is already used for training.
Why does it take a lot of data? Afaik you can select lower quality in youtube-dl but you don't even need video do you?
Earlier quoted context omitted.
Hi! What should you need to implement other language i.e. Italian or French? I mean: it's a problem due to the less of datas or what? Another question: could you use for example mozilla voice data to train/test?
Data and compute are the largest hurdles. I only have one GPU and training one model takes 3+ days, so I am limited by that. Also, scraping from YouTube takes time and a lot of storage (multiple TBs). Mozilla Common Voice data is already used for training.
Is there an open-source or paid SDK/API that I can use to create a group voice chat mobile app with "live" transcription? Or something that can plug-in to a system like this? I looked at Twilio but they seem to only offer a means to do it on their VOIP/SIP product.
Earlier quoted context omitted.
Why on earth would you put a demo front and center that shows your software doing a terrible, terrible job?
That's called setting up expectations. If you know your project might interest people but needs work, why pretends it's good when it's not? They seem to be courting contributors more than users anyway. I found the video to be funny. It nicely highlights both the current limitations and the ambition of the projects. Bold choice certainly but I think it works.
What datasets are used to train the models?
LibriSpeech, Tatoeba, Common Voice and scraped YouTube videos.
this would give you absolutely perfect sync down to the word, I assume... I don't know about the cost if you paid ratecard though, perhaps you can do some partnership with them since yours is a symetrical product
Is there an open-source or paid SDK/API that I can use to create a group voice chat mobile app with "live" transcription? Or something that can plug-in to a system like this? I looked at Twilio but they seem to only offer a means to do it on their VOIP/SIP product.
Agora also talk about this, but I haven't used it myself https://www.agora.io/en/
Earlier quoted context omitted.
Data and compute are the largest hurdles. I only have one GPU and training one model takes 3+ days, so I am limited by that. Also, scraping from YouTube takes time and a lot of storage (multiple TBs). Mozilla Common Voice data is already used for training.
Why does it take a lot of data? Afaik you can select lower quality in youtube-dl but you don't even need video do you?
But you need supervised data too.
Nice project and bold to demo with a French native speaking English. On a side project, I'm looking at the best interface to facilitate further edition (correction) of the recognized text. Target is local councils and regional parliament, where sessions are usually recorded but without transcripts. If xx% accuracy is enough to identify keywords, manual edition is still required to not distort precise meaning. Nothing…
I’m really interested in this project too. Been thinking about similar solutions for a while now. I looked into Kaldi and Mozilla Deep Speech but the former seems geared at ASR experts and the latter didn’t seem suited for my particular application (longer recorded audio or real time stream)
Earlier quoted context omitted.
Data and compute are the largest hurdles. I only have one GPU and training one model takes 3+ days, so I am limited by that. Also, scraping from YouTube takes time and a lot of storage (multiple TBs). Mozilla Common Voice data is already used for training.
For the compute problem: maybe you can use cloud server gpu powered as https://www.paperspace.com/ I don't know update prices but I remember it was quite affordable.
Relative to what? Paperspace is one of the costlier GPU providers.
Hey HN! I've been working on this for a while now. While there are other on-premise solutions using older models such as DeepSpeech [0], I haven't found a deployable project supporting multiple languages using the recent RNN-T Architecture [1]. Please note that this does not achieve SotA performance. Also, I've only trained it on one GPU so there might be room for improvement. Edit: Don't expect good performance :D t…