Automatic Spoken Language Identification using Convolutional Neural Networks
Resumo
Automatic Spoken Language Identification systems classify the spoken language automatically and can be used in many tasks, for example, to support Automatic Speech Recognition or Video Recommendation systems. In this work, we propose an automatic language identification model obtained through a Convolutional Neural Network trained over audio spectrograms on Portuguese, English and Spanish languages. The audio for the model training was obtained through audiobooks and different corpora for speech recognition systems. The audios were used to generate instances having five seconds each. We addressed the limitation of having few speakers in our dataset with simple data augmentation techniques such as speed and pitch changing on the original instances to increase the size of the dataset. The proposed model was optimized with a random hyperparameter search which provided a final model able to identify the proposed languages with 83% of accuracy on a new, unseen test data, made with audios from different sources.
Referências
Dehak, N., Torres-Carrasquillo, P. A., Reynolds, D., Dehak, R.: Language recognition via i-vectors and dimensionality reduction. In: Twelfth annual conference of the international speech communication association (2011).
Goodfellow, I., Bengio, Y., Courville, A.: Deep learning. MIT press (2016). [4] Hernández-Mena, C. D., Herrera-Camacho, J.: Ciempiess: A new open-sourced mexican spanish radio corpus. In: Proceedings of the ninth international conferenceon language resources and evaluation (LREC’14). pp. 371–375. European LanguageResources Association (ELRA) Reykjavik, Iceland (2014).
Li, H., Ma, B., Lee, C. H.: A vector space modeling approach to spoken language identification. IEEE Transactions on Audio, Speech, and Language Processing15(1), 271–284 (2006).
Montavon, G.: Deep learning for spoken language identification (01 2009).
Oponowicz, T.: Spoken language identification (2018), https://github.com/tomasz-oponowicz/spoken_language_identification.
Quintanilha, I. M., Biscainho, L. W. P., Netto, S.L.: Towards an end-to-end speech recognizer for portuguese using deep neural networks. XXXV Simpósio Brasileiro de Telecomunicações e Processamento de Sinais pp. 709–714 (2017).
Revay, S., Teschke, M.: Multiclass language identification using deep learning onspectral images of audio signals (2019).
Richardson, F., Reynolds, D., Dehak, N.: Deep neural network approaches to speaker and language recognition. IEEE signal processing letters22(10), 1671–1675 (2015).
Schramm, M., Freitas, L., Zanuz, A., Barone, D.: Cslu: Spoltech brazilian portuguese version 1.0 ldc2006s16 (2006).
Zazo, R., Lozano-Diez, A., Gonzalez-Dominguez, J., Toledano, D. T., Gonzalez-Rodriguez, J.: Language identification in short utterances using long short-term memory (lstm) recurrent neural networks. PloS one11(1), e0146917 (2016).
