Where can I find tests for accents?

[El_Smasho]

I know there are ways to test people’s accents to determine where they were born (assuming they didn’t spend formative years elsewhere).

Tests like:

• Do you pronounce the phrase “greaSy spoon” or “greaZy spoon”?
• Do you say these words alike: “Mary, merry, marry”?
• Do you say these words alike: “bird, burred”?

I’d like to find out how close they can get.

I don’t even know what this process or -olgy this is called.

[xash]

Perhaps it comes under Linguistic Anthropology, but I’m not sure.

Can’t answer your question directly, but this might help you in your research:

http://www.bl.uk/collections/sound-archive/accents.html

http://www.otago.ac.nz/anthropology/Linguistic/References/References.html

[El_Smasho]

Thanks. That’s a good place for me to start looking, all right.

[xash]

You could also look here:

http://directory.google.com/Top/Science/Social_Sciences/Anthropology/Linguistic_Anthropology/

And below are a couple of references to research on automatic accent identification systems.

From :

AUTOMATIC ACCENT IDENTIFICATION USING GAUSSIAN MIXTURE MODELS (note: .pdf file)

In this paper, we proposed a Gaussian mixture model (GMM) based Mandarin accent identification method. GMM method can avoid building model for phoneme or phoneme-class, which is not economic for many applications. Furthermore, GMM training is an unsupervised one: no transcriptions are needed, compared with the supervised HMM training. Other than the conventionally studied foreign accent, we explored the more challenging domestic accent identification problem. 4 typical Mandarin accent types are investigated. Experiments show that properly selected GMM parameter (number of components) can result in good identification performance.

We also investigated how many utterances per speaker are sufficient to identify his/her accent reliably. Usually more utterances will guarantee better accuracy, while in many applications we cannot obtain so much data. Fortunately, our experiments show that 3~5 utterances are good enough to be used to recognize a speaker’s accent by the proposed method. With 4 test utterances, about 11.7% and 15.5% error rate in accent classification was achieved for female and male speakers, respectively.

Finally, we showed the correlations among accents and provide some discussions. The accent identification system can be directly used to select accent-dependent model for speaker adaptation. Future work of its applications in speech recognition is undergoing.

And :

a577.pdf

Automatic identification of non-native accents is a difficult task, as illustrated by the results we have obtained with only 6 different accents. We have tested accent identification in isolation and integrated in a speech recognition system. The latter results have shown that it is preferable to train a recogniser with a mixture of accents.

The current work has used single pronunciation networks per word, a restriction which does not take into account the multiple ways in which aword can be pronouncedby a non-native speaker,depending on his reading competence and the differenced between his native phoneme inventory and the English one. We are currently working on deriving multiple pronunciation networks adapted to given accents and are also considering exploring rythmic cues in order to do a more effective accent identification.

[SkiChick33]

*Originally posted by El Smasho *
**I don’t even know what this process or -olgy this is called. **

It’s Dialectology. Really
http://www.google.com/search?sa=Google+Search&cat=&hl=en&q=Dialectology