I tried several times to get Google Translate’s voice recognition to properly recognize my spoken pronunciation of this sentence:
他在大學念政治學。
But 9 times out of 10, Google fails to recognize my pronunciation of 政治學. It seems to recognize it best if I speak the entire sentence robotically, with each syllable taking the same amount of time, and exaggerating (slowing down) my pronunciation of the high-to-low falling tones on 政 (zheng4) and 治 (zhi4). Even then, Google only occasionally recognizes the word.
If anyone here is using Google Translate (either the smartphone app or the web page), could you try speaking the above sentence into Google Translate and see if Google properly recognizes it?
I’m wondering if my pronunciation is at fault, if Google’s recognition is at fault, or both.
Thanks for the confirmation. I guess my tones are sloppy. It also seems that Google Translate requires that the pacing of the sentence be approximately correct for accurate recognition (i.e. if you insert too many pauses between words, that seems to negatively affect recognition accuracy).
On the smartphone app, if you click the upper-left “hamburger” menu and go Settings, then click Region, you can set the Chinese region setting to Taiwan, Traditional.
Not necessarily. I think Google translate voice dislikes non native speakers from my experience. Obviously Google’s language model is trained on native speakers and may pick up other features of their pronunciation other than the tone. In fact I’ve found it highly contextualised, so if you get something wrong anywhere in what you are saying it kind of falls in a heap.
I’d say the problem is possibly a lazy second tone on the xue. This is a pretty common problem for people learning Chinese. After two hard descending fourth tones, it’s hard to make the second tone rise up properly and it turns into a third tone. Maybe try emphasizing the rising of that second tone when you pronounce zhengzhixue.
Another problem may be the pronunciation of xue, another hard one for learners of Chinese. Some people tend to divide the u and the e sounds into two syllables, when it should be one. You need to try and keep that diphthong as tight as possible.
The [ʈʂ] seems to be the thing that I was getting wrong. Although I was taking pains to curl my tongue back for that authentic retroflex sound on the zh, I was voicing my “zh” sounds with a bit of buzzing in the throat. But then I remembered that one of my textbooks specifically said the vocal cords do not vibrate during the zh sound.
Making this change has improved the recognition rate when I say 政治學. I probably still have to work on the tones also, but at least now the recognition rate is somewhere around 50% whereas before it was like 10%.
FWIW, I just tried it with a neutral tone on the 学 and it had no problem. Not sure about why. It was also okay with my 2nd tone on that. I may have a more, um, Mainland-ized-ish accent since I’ve been out of Taiwan for some time now and the folks I interpret for are generally from there, not Taiwan.
I was just doing a listening comprehension exercise and tried dictating the answers. I found this one quite difficult to get the correct voice recognition.
