When talking to Chinese learners about building features to help them study Chinese, one suggestion tops the list: speech recognition. Unfortunately, it just doesn’t work. Read on to find out why.
Not Designed for Learners
Speech recognition is a “solved problem”. Software with speech recognition works for most people out of the box with zero training. And with a few minutes of training, it gets pretty accurate. The Chinese speech recognition software (a.k.a. speech-to-text) is also pretty mature. But It’s designed to understand native speakers, not learners of a language. Chinese learners don’t speak as well or accurately as Chinese natives. That’s what makes them learners.
Chinese learners come from all over
The second problem with speech recognition for Chinese: learners come from all over. Most learners speech Chinese with the accents: Americans sound American; Italians sound Italian; Koreans sound Korean. Speech recognition software would have to take into account all of these different accents. And to make things even more complicated, it would have to take understand regional accents as well. So it would need to understand a New Yorker’s accent, a Bostonian accent, the Texas twang, etc. Adding in all those accents would be a pretty long and expensive task.
Training Doesn’t Work
Training Chinese speech recognition wouldn’t work for individual Chinese learners. When people train speech recognition, they teach it their unique way of speaking. For a learner, that would be a disaster. If the speech engine learns from you, it will start accepting your incorrect pronunciation as correct. Since a learner doesn’t have proper pronunciation, it can’t be trained properly.
Can speech recognition ever be used to help people learn Chinese?
I think it’s possible. There are products on the market that do some interesting things, but they suffer from the problems I mentioned before. They also tend to fail in a few regards. A good speech recognition for Chinese learners needs to follow these rules:
- Never tell a learner they’re pronunciation was accurate when it’s not (no false positives)
- Always tell a learner when their pronunciation is correct
- Understand tones well
Speech recognition that doesn’t follow these rules will end up hurting learners by reinforcing their mistakes. Until there is speech recognition software that follows all of these rules, I’m going to stay away from it, and I suggest you do the same.
Does speech recognition software not work/work for you? Let me know why in the comments.