This contribution presents the development and evaluation of a spelled letter recognizer for automotive environments. Specifically, the spoken language dialog for the navigation system requires reliable recognition of thousands of city names. In this context the recognition of spelling sequences is needed as fall-back strategy and for the disambiguation of similar sounding names. For that purpose we have developed a speaker-independent spelled letter recognizer on the basis of hidden Markov models using the HTK toolkit. Speech data which have been collected in real-world driving situations are used for the training of the hidden Markov models. Several feature extraction schemes were investigated and compared with regard to the recognition performance of the system. The best results for both arbitrary spelling sequences and constrained city name recognition are achieved by a system with two-channel LDA and integrated noise reduction.
Cite as: Korthauer, A. (2001) Recognition of spelled city names in automotive environments. Proc. 7th European Conference on Speech Communication and Technology (Eurospeech 2001), 1581-1584, doi: 10.21437/Eurospeech.2001-385
@inproceedings{korthauer01_eurospeech,
author={Andreas Korthauer},
title={{Recognition of spelled city names in automotive environments}},
year=2001,
booktitle={Proc. 7th European Conference on Speech Communication and Technology (Eurospeech 2001)},
pages={1581--1584},
doi={10.21437/Eurospeech.2001-385}
}