roddy Posted May 13, 2015 at 01:45 PM Report Share Posted May 13, 2015 at 01:45 PM Wondering if there's anyone out there with any insight - there are plenty of language learning apps which will get you to repeat something into a microphone and tell you how you did - Babbel does this, Duolingo does, Rosetta Stone I believe does, etc. But I'm wondering how actually useful this is. Speech recognition is a lot better than it used to be - Siri, like, totally gets me. But speech recognition of a native speaker and speech recognition for second language learning - surely different kettles of fish? There are technologies out there designed for this purpose - for example Eduspeak says it has "patented human-calibrated pronunciation-scoring technology". Can you do that kind of processing on a phone or desktop, or do you have to run back to a server, like Siri does? Are there any apps out there that give you some kind of rating or feedback, as opposed to a yes / no response. Seems feasible that it could say 'your cat sounded more like cart' or 'the stress should be on the second syllable'. Is anyone doing that? This paper thinks it's possible, but I note that while Babbel had a scoring system in 2010, they now only have a green to go, red to repeat system. Quote Link to comment Share on other sites More sharing options...
Shelley Posted May 13, 2015 at 02:22 PM Report Share Posted May 13, 2015 at 02:22 PM The only type of thing like I to use is a spectrum analysis graph. So a native speaker speaks and generates a graph of the word and then you speak and below the native speaker your graph appears and you can compare. I like this visual way of doing it. My ear is not that good so it helps being visual. It also had a speedo type of graph to show how well you did from red (0) to green (100). I assume it compares the graphs, it doesn't actually need to recognise the words just compare the graphs. I figure that this is how speech recognition works, there is a data base of graphs for each word/sound and then there is a way of looking up and comparing, and I am guessing its the speed and accuracy of the looking up and comparing that has improved. i am no expert in this actual field but I work in audio electronics so I am just putting what I do know with what I have used. Quote Link to comment Share on other sites More sharing options...
imron Posted May 14, 2015 at 02:43 AM Report Share Posted May 14, 2015 at 02:43 AM Can you do that kind of processing on a phone or desktop, or do you have to run back to a server, like Siri does? I think you could, however there are caveats such as you'd be taking up a lot of processing power (bye, bye battery life) and also consuming large amounts of hard disk for sentence databases and samples for comparison, by sending everything to a server, you also have the benefit of being able to collect large amounts of data for further analysis and improvements to your processing engine. Quote Link to comment Share on other sites More sharing options...
markcarter Posted May 14, 2015 at 07:43 AM Report Share Posted May 14, 2015 at 07:43 AM I'm not sure about iOS, but on Android, there is built-in offline voice recognition for about 10 languages including Mandarin and Cantonese. Quote Link to comment Share on other sites More sharing options...
AdamD Posted May 14, 2015 at 07:59 AM Report Share Posted May 14, 2015 at 07:59 AM 出门问问 has always been pretty good at recognising my crappy spoken Chinese. If I ask for the nearest dumpling house or cat cafe, it delivers. http://chumenwenwen.com/ I presume that indicates something positive about the state of voice recognition for Chinese, but it doesn't do anything that would enhance my speech, like active feedback or a scoring system. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and select your username and password later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.