Publication Details
Topic identification of spoken documents using unsupervised acoustic unit discovery
Pappagari Raghavendra (FIT)
Ondel Lucas Antoine Francois, Mgr., Ph.D. (SSDIT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM)
Dehak Najim
Khudanpur Sanjeev
Černocký Jan, prof. Dr. Ing. (DCGM)
Gangashetty Suryakanth V (FIT)
topic identification, acoustic unit discovery, unsupervised learning, non-parametric Bayesian models
This paper investigates the application of unsupervised acoustic unit discovery for topic identification (topic ID) of spoken audio documents. The acoustic unit discovery method is based on a nonparametric Bayesian phone-loop model that segments a speech utterance into phone-like categories. The discovered phone-like (acoustic) units are further fed into the conventional topic ID framework. Using multilingual bottleneck features for the acoustic unit discovery, we show that the proposed method outperforms other systems that are based on cross-lingual phoneme recognizer.
This paper investigates the application of unsupervised acoustic unit discovery for topic identification (topic ID) of spoken audio documents. The acoustic unit discovery method is based on a nonparametric Bayesian phone-loop model that segments a speech utterance into phone-like categories. The discovered phone-like (acoustic) units are further fed into the conventional topic ID framework. Using multilingual bottleneck features for the acoustic unit discovery, we show that the proposed method outperforms other systems that are based on cross-lingual phoneme recognizer.
@inproceedings{BUT144450,
author="Santosh {Kesiraju} and Raghavendra {Pappagari} and Lucas Antoine Francois {Ondel} and Lukáš {Burget} and Najim {Dehak} and Sanjeev {Khudanpur} and Jan {Černocký} and Suryakanth V {Gangashetty}",
title="Topic identification of spoken documents using unsupervised acoustic unit discovery",
booktitle="Proceedings of ICASSP 2017",
year="2017",
pages="5745--5749",
publisher="IEEE Signal Processing Society",
address="New Orleans",
doi="10.1109/ICASSP.2017.7953257",
isbn="978-1-5090-4117-6",
url="https://www.fit.vut.cz/research/publication/11470/"
}