Automatic/Computational Language Detection in Speech

Question

Are there any packages that do much the same thing for audio/speech that the langid component/corpus of NLTK does for written text? The langid corpus/tool makes surprisingly accurate guesses about a language after training on popular trigram counts for a bunch of languages (from a web-scraped corpus), by comparing overlap of the new document with the trigrams for each known language.

I'm guessing that there are two really hard problems here:

1. Developing an audio segmenter/chunker (I expect there are decent solutions out there already for this)
2. Collecting a corpus of audio for many different languages

I don't have a whole lot of experience in speech recognition and that whole domain. Is this even feasible, or is speech way too hard to process for this purpose, given the state of voice recognition at this point?

Answer 1

Your task is called Spoken Language Identification (LID) or Language Recognition. The best current systems combine acoustic and phonotactic information sources.

• This paper was the first to show the superiority of GMMs with shifted delta cepstral coefficients:

  P. A. Torres-Carrasquillo, E. Singer, M. A. Kohler, R. J. Greene, D. A. Reynolds, J. R. Deller, Jr., Approaches to language identication using Gaussian mixture models and shifted delta cepstral features, in: International Conference on Spoken Language Processing, 2002, pp. 89-92. PDF

• This paper used SVM on top of GMM and it is considered the best method using acoustic features:

  William M. Campbell, Joseph P. Campbell, Douglas A. Reynolds, E. Singer, and P. A. Torres-Carrasquillo, “Support vector machines for speaker and language recognition.,” Computer Speech and Language, vol. 20, no. 2-3, pp. 210–229, 2006. PDF

• This paper used phonotactics and provides one of the best current results:

  Haizhou Li, Bin Ma, and Chin-Hui Lee, “A vector space modeling approach to spoken language identification,” IEEE Transactions on Audio, Speech and Language Processing, vol. 15, pp. 271–284, January 2007. PDF