Minimum Phone Error and I-smoothing for improved discriminative training

Povey, Daniel; Woodland, Philip C.

doi:10.1109/icassp.2002.5743665

Cited by 425 publications

(356 citation statements)

References 6 publications

Supporting

Mentioning

353

Contrasting

Order By: Relevance

“…Both systems extract Mel frequency cepstral coefficients (MFCCs) with standard normalization and adaptation techniques: cepstral vocal tract length normalization (VTLN), heteroscedastic linear discriminant analysis (HLDA), cepstral mean and variance normalization, and maximum likelihood linear regression (MLLR). Both systems have gender-dependent acoustic models trained discriminatively using variants of minimum phone error (MPE) training with maximum mutual information (MMI) priors (Povey and Woodland, 2002). Both train their acoustic models on approximately 2400 hours of conversational telephone speech from the Switchboard, CallHome and Fisher corpora, consisting of 360 hours of speech used in the 2003 evaluation plus 1820 hours of noisy "quick transcriptions" from the Fisher corpus, although with different segmentation and filtering.…”

Section: Datamentioning

confidence: 99%

Which words are hard to recognize? Prosodic, lexical, and disfluency factors that increase speech recognition error rates

Goldwater

Jurafsky

Manning

2010

Speech Communication

125

View full text Add to dashboard Cite

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. ACCEPTED MANUSCRIPT AbstractDespite years of speech recognition research, little is known about which words tend to be misrecognized and why. Previous work has shown that errors increase for infrequent words, short words, and very loud or fast speech, but many other presumed causes of error (e.g., nearby disfluencies, turn-initial words, phonetic neighborhood density) have never been carefully tested. The reasons for the huge differences found in error rates between speakers also remain largely mysterious.Using a mixed-effects regression model, we investigate these and other factors by analyzing the errors of two state-of-the-art recognizers on conversational speech. Words with higher error rates include those with extreme prosodic characteristics, those occurring turninitially or as discourse markers, and doubly confusable pairs: acoustically similar words that also have similar language model probabilities. Words preceding disfluent interruption points (first repetition tokens and words before fragments) also have higher error rates. Finally, even after accounting for other factors, speaker differences cause enormous variance in error rates, suggesting that speaker error rate vari ance is not fully explained by differences in word choice, fluency, or prosodic characteristics. We also propose that doubly confusable pairs, rather than high neighborhood density, may better explain phonetic neighborhood errors in human speech processing.

show abstract

Section: Datamentioning

confidence: 99%

Which words are hard to recognize? Prosodic, lexical, and disfluency factors that increase speech recognition error rates

Goldwater

Jurafsky

Manning

2010

Speech Communication

125

View full text Add to dashboard Cite

show abstract

“…It works in the same manner as the standard MAP, only the input HMM has to be discriminatively trained with the same objective function. For discriminative adaptation it is strongly recommended to use I-smoothing method to boost stability of new estimates [13].…”

Section: Frame-discriminative Adaptationmentioning

confidence: 99%

Discriminative Training of Gender-Dependent Acoustic Models

Vanĕk

Psutka

Zelinka

et al. 2009

Text, Speech and Dialogue

View full text Add to dashboard Cite

Abstract. The main goal of this paper is to explore the methods of genderdependent acoustic modeling that would take the possibly of imperfect function of a gender detector into consideration. Such methods will be beneficial in realtime recognition tasks (eg. real-time subtitling of meetings) when the automatic gender detection is delayed or incorrect. The goal is to minimize an impact to the correct function of the recognizer. The paper also describes a technique of unsupervised splitting of training data, which can improve gender-dependent acoustic models trained on the basis of manual markers (male/female). The idea of this approach is grounded on the fact that a significant amount of "masculine" female and "feminine" male voices occurring in training corpora and also on frequent errors in manual markers.

show abstract

“…To improve the generality of MBE training, the I-smoothing technique [10] is employed to provide better parameter estimates. This technique can be regarded as interpolating the MBE and ML auxiliary functions according to the amount of data available for each Gaussian mixture …”

Section: I-smoothing Updatementioning

confidence: 99%

“…The extended EM algorithm, which utilizes a weak-sense auxiliary function [9] and has been applied in the minimum phone error (MPE) discriminative training approach [10] for ASR, can be adapted to solve Eq.(6). …”

Section: Objective Function Optimization and Update Formulaementioning

confidence: 99%

A Minimum Boundary Error Framework for Automatic Phonetic Segmentation

Kuo

Wang

2006

Chinese Spoken Language Processing

View full text Add to dashboard Cite

Abstract. This paper presents a novel framework for HMM-based automatic phonetic segmentation that improves the accuracy of placing phone boundaries. In the framework, both training and segmentation approaches are proposed according to the minimum boundary error (MBE) criterion, which tries to minimize the expected boundary errors over a set of possible phonetic alignments. This framework is inspired by the recently proposed minimum phone error (MPE) training approach and the minimum Bayes risk decoding algorithm for automatic speech recognition. To evaluate the proposed MBE framework, we conduct automatic phonetic segmentation experiments on the TIMIT acoustic-phonetic continuous speech corpus. MBE segmentation with MBE-trained models can identify 80.53% of human-labeled phone boundaries within a tolerance of 10 ms, compared to 71.10% identified by conventional ML segmentation with ML-trained models. Moreover, by using the MBE framework, only 7.15% of automatically labeled phone boundaries have errors larger than 20 ms.

show abstract

Minimum Phone Error and I-smoothing for improved discriminative training

Cited by 425 publications

References 6 publications

Which words are hard to recognize? Prosodic, lexical, and disfluency factors that increase speech recognition error rates

Which words are hard to recognize? Prosodic, lexical, and disfluency factors that increase speech recognition error rates

Discriminative Training of Gender-Dependent Acoustic Models

A Minimum Boundary Error Framework for Automatic Phonetic Segmentation

Contact Info

Product

Resources

About