Incorporating label dependency for ASR error detection via RNN

Errattahi, Rahhal; Salmam, Fatima Zahra; Ouahmane, Hassan

doi:10.1016/j.procs.2019.01.069

Cited by 7 publications

(7 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Both, V-RNN and MLP models consist of a single layer of 2048 units with a relu [19] activation function as described in [8]. The ULSTM consists of 1 hidden layer of 2048 staked LSTM units.…”

Section: Methodsmentioning

confidence: 99%

“…So, the conditional independence assumption is not satisfied in this application. This fact motivated us to propose a new Variant of Recurrent Neural Network (V-RNN) as the classifier for ASR error detection for the first time in [8].…”

Section: Classifiersmentioning

confidence: 99%

“…To tackle these problems, we have been developing a new approach for ASR error detection and error type classification [3,7,8]. We have targeted a new and different scenario where information about the inner workings of the ASR system is not accessible.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Evaluation of the effectiveness and efficiency of state-of-the-art features and models for automatic speech recognition error detection

et al. 2021

Self Cite

View full text Add to dashboard Cite

Speech based human-machine interaction and natural language understanding applications have seen a rapid development and wide adoption over the last few decades. This has led to a proliferation of studies that investigate Error detection and classification in Automatic Speech Recognition (ASR) systems. However, different data sets and evaluation protocols are used, making direct comparisons of the proposed approaches (e.g. features and models) difficult. In this paper we perform an extensive evaluation of the effectiveness and efficiency of state-of-the-art approaches in a unified framework for both errors detection and errors type classification. We make three primary contributions throughout this paper: (1) we have compared our Variant Recurrent Neural Network (V-RNN) model with three other state-of-the-art neural based models, and have shown that the V-RNN model is the most effective classifier for ASR error detection in term of accuracy and speed, (2) we have compared four features’ settings, corresponding to different categories of predictor features and have shown that the generic features are particularly suitable for real-time ASR error detection applications, and (3) we have looked at the post generalization ability of our error detection framework and performed a detailed post detection analysis in order to perceive the recognition errors that are difficult to detect.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Classifiersmentioning

confidence: 99%

See 1 more Smart Citation

Evaluation of the effectiveness and efficiency of state-of-the-art features and models for automatic speech recognition error detection

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

“…The recurrent neural network (RNN) deep-learning algorithm is a special neural network structure that uses the partial derivative of the loss function to adjust the weight of each unit and shares the parameters in the connection of each step of input, output, and hidden states. This sharing mechanism can significantly lower the complexity of the model, shorten the training time, and maintain high accuracy [23][24][25][26]; hence, it has greater advantages over other ANNs. Additionally, in the current CLQ evaluation research, many studies have focused only on evaluating a single aspect of CLQ (such as the evaluation of natural grades) while ignoring the multifunctional evaluation of CLQ (such as the evaluation of utilization grades and economic grades).…”

Section: Introductionmentioning

confidence: 99%

Cultivated Land Quality Evaluated Using the RNN Algorithm Based on Multisource Data

Zhou

Zhao

et al. 2022

Remote Sensing

View full text Add to dashboard Cite

Cultivated land quality (CLQ) is associated with national food security, benign economic development, social harmony, and stability. The scientific evaluation of CLQ provides the basis for achieving the “trinity” protection of cultivated land quantity, and quality, as well as ecology. However, the current research on CLQ evaluation has some limitations, mainly the poor consideration of evaluation indicators, time-consuming and labor-intensive data acquisition, and low precision of evaluation at the regional scale. Therefore, this study introduced multisource data to evaluate CLQ and proposed a new method for CLQ evaluation (natural grade evaluation, utilization grade evaluation, and economic grade evaluation), combining multisource data and the recurrent neural network (RNN) algorithm. Initially, optimal indicators were determined by correlation analysis and generalized linear regression coefficient methods based on factors related to CLQ acquired from multisource data. Then, CLQ evaluation models were constructed with the RNN algorithm on the basis of the aforementioned optimal indicators. Finally, the models were adopted to map CLQ. The present study was carried out in Guangzhou City, Guangdong Province, China. According to the results: (1) CLQ showed close relationship to pH, effective soil layer thickness (EST), chemical fertilizer application rate (CHFE), organic matter content (OMC), annual accumulated temperature (TEMA), 5–15 cm soil depth soil cation exchange capacity (CEC515), 0–5 cm soil depth soil cation exchange capacity (CEC05), 5–15 cm soil depth soil organic carbon content (SOC515), 0–5 cm soil depth soil organic carbon content (SOC05), field slope (FS), groundwater level (GWL), and terrain slope (TS). (2) All modeling accuracies (R2) were greater than 0.80 for the CLQ evaluation models constructed based on the RNN algorithm. The area and spatial distribution of each grade of CLQ evaluation were consistent with the actual situation. The best and the worst quality cultivated land occupied a small area, and the area without a gap with the actual CLQ was as high as 76%, indicating that the model results were reliable. The study shows the suitability of the method for evaluating CLQ at the regional scale, offering a scientific foundation for the rational utilization and management of cultivated land resources, as well as a reference for evaluating CLQ in the future.

show abstract

“…Our objective to promote the correct ASR hypotheses for unlabeled speech data is similar to Tong et al's, however we focus on limited data scenarios and we use an error detector to bias the supervision towards correct words in both matched and mismatched scenarios. Modern ASR error detectors [13][14][15] are more powerful than the lattice posteriorbased confidence scores used to weight per-frame gradients [6] or to discard erroneous words [16,17] or utterances [18,19] in most semi-supervised neural AM training studies. Yet, to the best of our knowledge, they have not been leveraged for semisupervised LF-MMI training so far.…”

Section: Introductionmentioning

confidence: 99%

On Semi-Supervised LF-MMI Training of Acoustic Models with Limited Data

Sheikh¹,

Illina²

2020

Interspeech 2020

View full text Add to dashboard Cite

This work investigates semi-supervised training of acoustic models (AM) with the lattice-free maximum mutual information (LF-MMI) objective in practically relevant scenarios with a limited amount of labeled in-domain data. An error detection driven semi-supervised AM training approach is proposed, in which an error detector controls the hypothesized transcriptions or lattices used as LF-MMI training targets on additional unlabeled data. Under this approach, our first method uses a single error-tagged hypothesis whereas our second method uses a modified supervision lattice. These methods are evaluated and compared with existing semi-supervised AM training methods in three different matched or mismatched, limited data setups. Word error recovery rates of 28 to 89% are reported.

show abstract

Incorporating label dependency for ASR error detection via RNN

Cited by 7 publications

References 13 publications

Evaluation of the effectiveness and efficiency of state-of-the-art features and models for automatic speech recognition error detection

Evaluation of the effectiveness and efficiency of state-of-the-art features and models for automatic speech recognition error detection

Cultivated Land Quality Evaluated Using the RNN Algorithm Based on Multisource Data

On Semi-Supervised LF-MMI Training of Acoustic Models with Limited Data

Contact Info

Product

Resources

About