Improved Recognition of Spontaneous Hungarian Speech—Morphological and Acoustic Modeling Techniques for a Less Resourced Task

Mihajlik, Péter; Tüske, Zoltán; Tarján, Balázs; Németh, Bottyán; Fegyó, Tibor

doi:10.1109/tasl.2009.2038807

Cited by 26 publications

(18 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Building language models for morphologically rich languages (like Hungarian) is a challenging task due to data sparseness, high OOV rate and large lexicons [15]. These issues are usually handled by using subword language models instead of the standard word-based ones ( [16], [17], [18]). However, in our former studies we found that the benefit from using subword models highly depends on the amount of available training data and the speech genre of the recognition task ( [19], [20], [21]).…”

Section: Language Modelingmentioning

confidence: 99%

Automated transcription of conversational Call Center speech – with respect to non-verbal acoustic events

Sárosi

Tarján

Fegyó

et al. 2014

IDT

View full text Add to dashboard Cite

This paper summarizes our recent efforts made to transcribe real-life Call Center conversations automatically with respect to non-verbal acoustic events, as well. Future Call Centers-as cognitive infocom systems-must respond automatically not only for well formed utterances but also for spontaneous and non-word speaker manifestations and must be robust against sudden noises. Conversational telephony speech transcription itself is a big challenge, primarily we address this issue on real-life (Bank and Insurance) tasks. In addition, we introduce several non-word acoustic modeling approaches and their integration to LVCSR (Large Vocabulary Continuous Speech Recognition). In the experiments, one and two channel (client and agent speech merged into one or left in two separate audio stream) transcription results, cross-task results and the handling of transcription data insufficiency are investigated-in parallel with the non-verbal acoustic event modeling. On the agent side less than 15% word error rate could be achieved and the best error rate reduction is 20% (relative) due to the inclusion of various written corpora and due to acoustic event handling.

show abstract

Section: Language Modelingmentioning

confidence: 99%

Automated transcription of conversational Call Center speech – with respect to non-verbal acoustic events

Sárosi

Tarján

Fegyó

et al. 2014

IDT

View full text Add to dashboard Cite

show abstract

“…All of these three effects -varying word order, large vocabulary and spontaneity -hamper statistic models' ability to yield consistent estimates by high confidence. Since data sparsity issues can be often handled by estimating language models on statically derived subword units (morphs) [5,9,10] in morphologically rich languages, we extended our investigation to morph-based language models, as well.…”

Section: Introductionmentioning

confidence: 99%

“…In [7], subword RNNLMs were trained on Finnish and Estonian conversations and used for rescoring lattices generated with conventional back-off models. Subword language models have already been applied successfully for recognition of Hungarian conversational speech [10,19], but neural language models have not been used before to the best of our knowledge. We found only one mention of application of morph-based approximated RNNLMs in the first pass of an ASR system [13], however this paper did not provide morph-based ASR results.…”

Section: Introductionmentioning

confidence: 99%

Investigation on N-Gram Approximated RNNLMs for Recognition of Morphologically Rich Speech

Tarján

Szaszák

Fegyó

et al. 2019

Statistical Language and Speech Processing

View full text Add to dashboard Cite

Recognition of Hungarian conversational telephone speech is challenging due to the informal style and morphological richness of the language. Recurrent Neural Network Language Model (RNNLM) can provide remedy for the high perplexity of the task; however, twopass decoding introduces a considerable processing delay. In order to eliminate this delay we investigate approaches aiming at the complexity reduction of RNNLM, while preserving its accuracy. We compare the performance of conventional back-off n-gram language models (BNLM), BNLM approximation of RNNLMs (RNN-BNLM) and RNN n-grams in terms of perplexity and word error rate (WER). Morphological richness is often addressed by using statistically derived subwords -morphs -in the language models, hence our investigations are extended to morph-based models, as well. We found that using RNN-BNLMs 40% of the RNNLM perplexity reduction can be recovered, which is roughly equal to the performance of a RNN 4-gram model. Combining morph-based modeling and approximation of RNNLM, we were able to achieve 8% relative WER reduction and preserve real-time operation of our conversational telephone speech recognition system.

show abstract

“…For a comprehensive survey of different methods, see [2]. The subword vocabularies selected using unsupervised machine learning methods have been shown to perform well [3,4], so neither a morphological analyzer nor an annotated training corpus is required. One common unsupervised method is Morfessor [5], which is used as a baseline method in this work.…”

Section: Introductionmentioning

confidence: 99%

Learning a subword vocabulary based on unigram likelihood

Varjokallio

Kurimo

Virpioja

2013

2013 IEEE Workshop on Automatic Speech Recognition and Understanding

View full text Add to dashboard Cite

Using words as vocabulary units for tasks like speech recognition is infeasible for many morphologically rich languages, including Finnish. Thus, subword units are commonly used for language modeling. This work presents a novel algorithm for creating a subword vocabulary, based on the unigram likelihood of a text corpus. The method is evaluated with entropy measure and a Finnish LVCSR task. Unigram entropy of the text corpus is shown to be a good indicator for the quality of higher order n-gram models, also resulting in high speech recognition accuracy.

show abstract

Improved Recognition of Spontaneous Hungarian Speech—Morphological and Acoustic Modeling Techniques for a Less Resourced Task

Cited by 26 publications

References 30 publications

Automated transcription of conversational Call Center speech – with respect to non-verbal acoustic events

Automated transcription of conversational Call Center speech – with respect to non-verbal acoustic events

Investigation on N-Gram Approximated RNNLMs for Recognition of Morphologically Rich Speech

Learning a subword vocabulary based on unigram likelihood

Contact Info

Product

Resources

About