Light supervision in acoustic model training

Nguyen, Long; Xiang, Bing

doi:10.1109/icassp.2004.1325953

Cited by 40 publications

(54 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For each configuration, we run the simulation models until we get enough training data such that the learned policies on that corpus do not change anymore (40,000 dialogs are generated by each model). After that, the learned new policies are implemented into the original system respectively 3 . Finally, we use our most realistic model, the PM, to interact with each new system 500 times to evaluate the new systems' performances.…”

Section: Methodsmentioning

confidence: 99%

Automatic acquisition of grammatical types for nouns

Bel

Espeja

Marimon

2007

Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics

View full text Add to dashboard Cite

The work 1 we present here is concerned with the acquisition of deep grammatical information for nouns in Spanish. The aim is to build a learner that can handle noise, but, more interestingly, that is able to overcome the problem of sparse data, especially important in the case of nouns. We have based our work on two main points. Firstly, we have used distributional evidences as features. Secondly, we made the learner deal with all occurrences of a word as a single complex unit. The obtained results show that grammatical features of nouns is a level of generalization that can be successfully approached with a Decision Tree learner.

show abstract

Section: Methodsmentioning

confidence: 99%

Automatic acquisition of grammatical types for nouns

Bel

Espeja

Marimon

2007

Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics

View full text Add to dashboard Cite

show abstract

“…Erroneous patterns in C3 will be rejected by the second classifier. Note that the conventional method [10], [11] simply accepts C1 and C2, but our proposed method can also incorporate more effective data (C4 and C5) and remove erroneous data (C2).…”

Section: Cascaded Classifiers For Word-level Data Selectionmentioning

confidence: 99%

“…Note that the conventional method [10], [11] is equivalent to simply using C1 and C2. The objective of this study is to incorporate more effective data (C4 and C5) while removing erroneous data (C2 and C3).…”

Section: Category Of Word Alignment Patternsmentioning

confidence: 99%

“…• Conventional: the conventional lightly supervised training method which selects the data based on simple matching of the ASR hypothesis and the caption text [10], [11] (upper part of Fig. 3).…”

Section: Asr Performance With Enhanced Model Trainingmentioning

confidence: 99%

“…However, in the case of TV programs, closed caption texts are not so verbatim because of the space constraint, and thus can be used in an indirect manner for lightly supervised training. A typical method [10], [11] consists of two steps. In the first step, a biased language model is constructed based on the closed caption text of the relevant program to guide the baseline ASR system to decode the audio content.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Automatic Lecture Transcription Based on Discriminative Data Selection for Lightly Supervised Acoustic Model Training

Akita

Kawahara

2015

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYThe paper addresses a scheme of lightly supervised training of an acoustic model, which exploits a large amount of data with closed caption texts but not faithful transcripts. In the proposed scheme, a sequence of the closed caption text and that of the ASR hypothesis by the baseline system are aligned. Then, a set of dedicated classifiers is designed and trained to select the correct one among them or reject both. It is demonstrated that the classifiers can effectively filter the usable data for acoustic model training. The scheme realizes automatic training of the acoustic model with an increased amount of data. A significant improvement in the ASR accuracy is achieved from the baseline system and also in comparison with the conventional method of lightly supervised training based on simple matching.

show abstract

Step-by-step and integrated approaches in broadcast news speaker diarization

Meignier

Moraru

Fredouille

et al. 2006

Computer Speech & Language

114

View full text Add to dashboard Cite

International audienceThis paper summarizes the collaboration of the LIA and CLIPS laboratories on speaker diarization of broadcast news during the spring NIST Rich Transcription 2003 evaluation campaign (NIST-RTÕ03S). The speaker diarization task consists of segmenting a conversation into homogeneous segments which are then grouped into speaker classes. Two approaches are described and compared for speaker diarization. The first one relies on a classical two-step speaker diarization strategy based on a detection of speaker turns followed by a clustering process, while the second one uses an integrated strategy where both segment boundaries and speaker tying of the segments are extracted simultaneously and challenged during the whole process. These two methods are used to investigate various strategies for the fusion of diarization results. Furthermore, segmentation into acoustic macro-classes is proposed and evaluated as a priori step to speaker diarization. The objective is to take advantage of the a priori acoustic information in the diariza-tion process. Along with enriching the resulting segmentation with information about speaker gender

show abstract

Light supervision in acoustic model training

Cited by 40 publications

References 9 publications

Automatic acquisition of grammatical types for nouns

Automatic acquisition of grammatical types for nouns

Automatic Lecture Transcription Based on Discriminative Data Selection for Lightly Supervised Acoustic Model Training

Step-by-step and integrated approaches in broadcast news speaker diarization

Contact Info

Product

Resources

About