Multilingual large vocabulary speech recognition: the European SQALE project

Young, Steve; Adda-Dekker, M.; Aubert, Xavier L.; Dugast, Christian; Gauvain, Jean‐Luc; Kershaw, Dan J.; Lamel, Lori; Leeuwen, David A. van; Pye, David; Robinson, Anthony J.; Steeneken, Herman J. M.; Woodland, Philip C.

doi:10.1006/csla.1996.0023

Cited by 39 publications

(12 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recognition systems developed originally for one language have been successfully ported to several languages, including systems developed by IBM (Cohen et al, 1997), Dragon (Barnett et al, 1996), BBN (Billa et al, 1997), Cambridge (Young et al, 1997), Philips (Dugast et al, 1995), MIT (Glass et al, 1995), and LIMSI (Lamel et al, 1995). The transformation of English systems to such diverse languages like German, Japanese, French, and Mandarin Chinese illustrates that speech technology generalizes across languages and that similar modeling assumptions hold for various languages.…”

Section: Introductionmentioning

confidence: 99%

Language-independent and language-adaptive acoustic modeling for speech recognition

Schultz

Waibel

2001

Speech Communication

328

222

View full text Add to dashboard Cite

With the distribution of speech technology products all over the world, the portability to new target languages becomes a practical concern. As a consequence our research focuses on the question of how to port LVCSR systems in a fast and efficient way. More specifically we want to estimate acoustic models for a new target language using speech data from varied source languages, but only limited data from the target language. For this purpose we introduce different methods for multilingual acoustic model combination and a polyphone decision tree specialization procedure. Recognition results using language dependent, independent and language adaptive acoustic models are presented and discussed in the framework of our GlobalPhone project which investigates LVCSR systems in 15 languages.

show abstract

Section: Introductionmentioning

confidence: 99%

Language-independent and language-adaptive acoustic modeling for speech recognition

Schultz

Waibel

2001

Speech Communication

328

222

View full text Add to dashboard Cite

show abstract

“…The use of perplexity also has an inherent difficulty in that it ignores the effects from acoustic confusability between vocabulary items. These issues are of particular significance for cross-task and cross-language evaluation (Young et al, 1997).…”

Section: Datamentioning

confidence: 99%

“…Each of these would have to be spelled correctly in the recognizer output in order to be judged as correct. In the European SQALE project (Young et al, 1997) the very high homophone rate in French was an important issue.…”

Section: Adapting Transcription and Scoring Practicesmentioning

confidence: 99%

“…If compounds are treated as distinct lexical entries, then meaningful comparisons are really only possible if all systems use a common lexicon. This approach was taken in the SQALE project but at the cost of a relatively poor lexical coverage and consequent high out-of-vocabulary rate (Young et al, 1997). The more general approach taken in the German CH evaluation was to calculate WER at the component word level while requiring that all components of the compound be present in some form in the recognizer output, whether composed into the target compound or not.…”

Section: Adapting Transcription and Scoring Practicesmentioning

confidence: 99%

See 1 more Smart Citation

Speech recognition evaluation: a review of the U.S. CSR and LVCSR programmes

Young

Chase

1998

Computer Speech & Language

View full text Add to dashboard Cite

“…The NIST evaluation campaigns were so successful that researchers in Europe followed the good example of the US and held their own evaluations of speech technology. One such evaluation was the EU-funded project SQALE 1 [24], in which large vocabulary speech recognition systems (20-65k words) were tested in British and American English, French and German, using read speech. Later, the French Technolangue program encompassed Evalda, the evaluation of many different human language technologies, among which the ESTER 2 evaluation for Broadcast News speech.…”

Section: Introductionmentioning

confidence: 99%

N-Best 2008: A Benchmark Evaluation for Large Vocabulary Speech Recognition in Dutch

Leeuwen¹

2012

Theory and Applications of Natural Language Processing

View full text Add to dashboard Cite

Automatic Speech Recognition (ASR) is a discipline of engineering that benefits particularly well from formal evaluations. There are several reasons for this. Firstly, speech recognition is basically a pattern recognition task, and to scientifically show that the system works it needs to be tested on fresh material that has never been observed by the system, or indeed the researchers themselves. This means that speech material for testing purposes needs to be collected, which requires quite some effort, but can formally only be used once. It is therefore more efficient if the evaluation material is used to determine the performance of several systems simultaneously, which suggests a common form of this kind of performance benchmarking: that of a formal evaluation. Secondly, after a system evaluation the evaluation material and protocol can be used for future researchers as a benchmark test: algorithms can be developed and tuned to increase performance on the test. By using a well-established formal evaluation protocol performance figures can directly be compared amongst different researchers in the literature, which gives more meaning to the actual figures. Thirdly, a benchmark evaluation gives researchers a clear focus and goal, and appears to stimulate the different research groups to get the best out of their system in a friendly competitive way.Formal evaluations in speech technology have their origin in the early 1990s of the last century, when the US Advanced Research Projects Agency (ARPA) organised regular evaluations in speech recognition executed by the National Institute of Standards and Technology (NIST) [16], soon followed by speaker [12] and language [13] recognition. In the early years the language of interest for speech recognition invariably was English, but as tasks got harder and performance got

show abstract

Multilingual large vocabulary speech recognition: the European SQALE project

Cited by 39 publications

References 9 publications

Language-independent and language-adaptive acoustic modeling for speech recognition

Language-independent and language-adaptive acoustic modeling for speech recognition

Speech recognition evaluation: a review of the U.S. CSR and LVCSR programmes

N-Best 2008: A Benchmark Evaluation for Large Vocabulary Speech Recognition in Dutch

Contact Info

Product

Resources

About