71–80 of 172 posts
Do they use template databases for different topics? (like afaik WolframAlpha) Do they use (scientific) ontologies or are the templates more flat and stored in a SQL database? [and web search results as fallback]
Earlier quoted context omitted.
Actually I think most of the work has been done in academia --- certainly that's where the recent deep learning stuff has come from. So, I don't think the important stuff is patented. (In general it's very hard to lock down ML improvements under patent. Once we can do it one way, and understand a little bit about what's working, we can usually replicate that performance with another technique.) The big problem for op…
> The big problem for open source speech recognition is training data. What exactly is needed for training - audio recordings with transcripts, human validation of recognized text? There are successful crowdsourced efforts for proofreading of OCR'ed text. Archive.org could host a CC-licensed archive of sound & transcripts. Recognition of the human voice is almost like writing, hopefully everyone could have access. Ed…
That is a common size for LVCSR, and you need something around that area to get good performance (maybe minimum 100h). In academic papers by Google, they usually use their own private training data set, with e.g. 1900h. (E.g.: http://arxiv.org/pdf/1402.1128.pdf)
Some crowdsourced effort to collect transcribed audio under a CC-licence would be great!
A while back I created a toy IRC bot that answers questions like this just by searching reddit and taking the top comment. It works surprisingly well for questions that are likely to have been asked before. And when it does work, the result is much better than a dry wikipedia excerpt. I then added some simple machine learning to filter search results for the most relevant threads which improved it quite a bit.
Who owns the key patents for voice recognition - AT&T, Nuance, IBM? When will they be expire and be available for use in open-source speech recognition? Edit: some data at http://www.quora.com/Is-anyone-working-on-an-open-source-ver... & https://news.ycombinator.com/item?id=4987875
You can even download trained models: http://kaldi-asr.org/
It supports many state-of-the-art methods, like DNNs, sequence training, etc. So you can get quite good results with it. To train it yourself, of course you need some good training data from somewhere.
Earlier quoted context omitted.
Try it on Wolfram Alpha. Actually, they all seem to me pretty limited compared to the answers you get from Wolfram Alpha. Also c.f. the responses to one of their test questions, "how much is a quarter cup of butter?". Google makes fun of the inquiry. Wolfram Alpha gives you a thorough nutritional profile, and links to variations based on international cup sizing and different types of butter.
Wolfram Alpha is amazing at discerning the intention of the question. In the article, the question "How old is the Lincoln Tunnel" struck me as incorrectly formatted for the parser (I know, that's the point), so I asked Siri, "When was the Lincoln Tunnel built." The Wikipedia article on the Lincoln Tunnel was returned. Wolfram Alpha was listed under other sources, so I chose that. The response? "1937"
Luckily if you say "Wolfram XXX" instead of just "XXX" Siri will route your question straight to Alpha no-questions-asked.
How do such personal agents work? I mean after the speech recognition and NLP (part-of-speech tagging, named-entity recognition, etc): Do they use template databases for different topics? (like afaik WolframAlpha) Do they use (scientific) ontologies or are the templates more flat and stored in a SQL database? [and web search results as fallback]
What do you mean by a template database?
Oh awesome, let's see... how many atoms are there in the universe? "The number of atoms in the entire observable universe is estimated to be within the range of 1078 to 1082." Thank you Google! http://www.google.com/search?q=how+many+atoms+are+in+the+uni... ]
Try it on Wolfram Alpha. Actually, they all seem to me pretty limited compared to the answers you get from Wolfram Alpha. Also c.f. the responses to one of their test questions, "how much is a quarter cup of butter?". Google makes fun of the inquiry. Wolfram Alpha gives you a thorough nutritional profile, and links to variations based on international cup sizing and different types of butter.
Earlier quoted context omitted.
50% is good enough if its done by Apple. 90% not good enough if its not done by Apple. This is how I interpret most responses when people talk about Apple.
Well, as accurate as I feel your statement is, I'd also say sometimes it depends on the 10% missing from the products not done by Apple. If it's an important 10% that Apple has nailed (or close enough) then sometimes the 50% is more valuable than the 90%. But we all have our biases as well. (Even as an Apple user I fully expect Google to always win the Knowledge Vault race. Its in their wheelhouse more so than Apple'…
Heh, you not only proved kumarm's point about apple fanboi'sm but served the proof on a silver platter with a little side of dessert.
Oh awesome, let's see... how many atoms are there in the universe? "The number of atoms in the entire observable universe is estimated to be within the range of 1078 to 1082." Thank you Google! http://www.google.com/search?q=how+many+atoms+are+in+the+uni... ]