| 1 |
Some issues affecting the transcription of hungarian broadcast audio
|
|
|
|
In: Annual Conference of the International Speech Communication Association ; https://hal.archives-ouvertes.fr/hal-01843430 ; Annual Conference of the International Speech Communication Association , Aug 2013, Lyon, France (2013)
|
|
Abstract:
International audience ; This paper reports on a speech-to-text (STT) transcription system for Hungarian broadcast audio developed for the 2012 Quaero evaluations. For this evaluation, no manually transcribed audio data were provided for model training, however a small amount of development data were provided to assess system performance. As a consequence, the acoustic models were developed in an unsupervised manner, with the only supervision provided indirectly by the language model. The language models were trained on texts downloaded from various websites, also without any speech transcripts. This contrasts with other STT systems for Hungarian broadcast audio which use at least 10 to 50 hours of manually transcribed data for acoustic training, and typically include speech transcripts in the language models. Based on mixed results previously reported applying morph-based approaches to agglutinative languages such as Hungarian, word-based language models were used. The initial Word Error Rate (WER) of the system using context-independent seed models from other languages of 59.8% on the 3h development corpus was reduced to 25.0% after successive training iterations and system refinement. The same system obtained a WER of 23.3% on the independent Quaero 2012 evaluation corpus (a mix of broadcast news and broadcast conversation data). These results compare well with previously reported systems on similar data. Various issues affecting system performance are discussed, such as amount of training data, the acoustic features and choice of text sources for language model training.
|
|
Keyword:
[INFO.INFO-CL]Computer Science [cs]/Computation and Language [cs.CL]; [INFO]Computer Science [cs]; agglutinative languages; Bottleneck MLP features; broadcast news transcription; Hungarian language; Large vocabulary continuous speech recognition (LVCSR); unsupervised training
|
|
URL: https://hal.archives-ouvertes.fr/hal-01843430
|
|
BASE
|
|
Hide details
|
|
|
| 2 |
Structured output layer neural network language models for speech recognition
|
|
|
|
In: ISSN: 2329-9290 ; EISSN: 2329-9304 ; IEEE/ACM Transactions on Audio, Speech and Language Processing ; https://hal.archives-ouvertes.fr/hal-01908377 ; IEEE/ACM Transactions on Audio, Speech and Language Processing, Institute of Electrical and Electronics Engineers, 2013, 21, pp.197-206 (2013)
|
|
BASE
|
|
Show details
|
|
|
| 3 |
Acoustic unit discovery and pronunciation generation from a grapheme-based lexicon
|
|
|
|
In: IEEE Automatic Speech Recognition and Understanding Workshop ; https://hal.archives-ouvertes.fr/hal-01843433 ; IEEE Automatic Speech Recognition and Understanding Workshop, Dec 2013, Olomouc, Czech Republic (2013)
|
|
BASE
|
|
Show details
|
|
|
|
|
