Live Streaming Speech Recognition Using Deep Bidirectional LSTM Acoustic Models and Interpolated Language Models

Jorge, Javier; Giménez, Adrià; Silvestre-Cerdà, Joan Albert; Civera, Jorge; Sanchís, Alberto; Juan, Alfons

doi:10.1109/taslp.2021.3133216

Cited by 15 publications

(12 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Amateur programmers and developers use question and answer websites like Stack Overflow for help during programming [12]. To assist Stack Overflow readers, researchers have proposed a range of proposals, including automatic tag suggestions and question templates for users asking questions [31], API abuse warnings [23], and API descriptions [28]. With the exception of preexisting synonymous tag pairs offered by Stack Overflow Beyer et al suggested a tool to provide synonymous tags for an input tag.…”

Section: Related Workmentioning

confidence: 99%

“…Example Check is a plugin for detecting API calls that are utilized incorrectly in stack overflow code snippets. It also shows how to utilise APIs correctly, including examples to help users learn how to use them properly [23]. It is proposed that StackDoc be used to enhance Stack Overflow with descriptions and examples of the Java APIs utilized in the inquiries [28].…”

Section: Related Workmentioning

confidence: 99%

“…The Word Embedding4 refers to a set of language modelling techniques and procedures used in natural language processing to map words or phrases in a dictionary to real-number vectors The Neural Network Language Model (NNLM) [23], [24], Latent Semantic Analysis (LSA) [25], and the Latent Dirichlet Distribution (LDA) [26] are some of the earlier advancements in word embedding. Word2Vec has recently gained popularity.…”

Section: The Word Embeddingmentioning

confidence: 99%

See 2 more Smart Citations

comprehensive study based on societal question and answer system using natural language processing with topic modelling for job assistantship in the world industry

Latha

Rao³

2022

ijhs

View full text Add to dashboard Cite

This breakthrough study explores public opinion on technical words or topics using data from social media. Individuals seeking work assistance must consider public opinion on the existence of terminological or technical obstacles, as well as their impact on the environment and society. Public assistance is also required for the execution of legislation and mitigation programmes. A public opinion study is required to better understand the social environment and social processes. Researchers are paying special attention to social media data because it provides extremely significant information on public perceptions and responses to conflicting socio-technical terminology or issues, such as quorums, stack overflows, and yahoo! It also responds to Twitter and other social media platforms, and it's frequently used to monitor and analyze how society reacts to natural and social calamities. Search for keywords or topics in topic models to identify distinct topics to get the most out of social media data. In classic topic models, users may provide an erroneous number of subjects, resulting in unsatisfactory grouping results. For data retrieval and cluster trend identification in these scenarios, exact representations are essential.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: The Word Embeddingmentioning

confidence: 99%

See 1 more Smart Citation

comprehensive study based on societal question and answer system using natural language processing with topic modelling for job assistantship in the world industry

Latha

Rao³

2022

ijhs

View full text Add to dashboard Cite

show abstract

“…Besides, one related critical on-line recognition technology is ASR which has extensively applied the adaption of LSTM networks. An optimized, low-latency streaming decoder built with bidirectional LSTM acoustic models, together with general interpolated language models published recently in Jorge et al 22…”

Section: Related Workmentioning

confidence: 99%

Apply an optimized NN model to low-dimensional format speech recognition and exploring the performance with restricted factors

Chen

Yeh

2022

Measurement and Control

View full text Add to dashboard Cite

The SCD (speech control detection) have received a lot of attention in recent years. A framework established by employing DNN-LSTM (deep neural network-long-short term memory) model for speech and text recognition is implemented in the current article. The performance of the build framework is analyzed with many different merits which consider many features, such as (with and without) noise, track number of speeches (ST (single track) and DT (double track)), and dropout ratio of data training. On the other hand, the speech discriminator model is developed and implemented with the DNN-LSTM framework, and the data sets are collected by four different persons. The adopted model performance is evaluated using the four different datasets, and each with 400–5000 training times. There are three parameters considered as the dominators for the performance evaluation of the completed speech platform. The results from the experiment with DT channel case clearly show that it outperforms the case with ST channel. It can see that the accuracy of the DNN-LSTM model increases from 0.3339 to 0.9696 and the loss rate decreases from 1.09984 to 0.19298 after adjusting the dropout ratio during the training step. This shows that the dropout ratio also dominates the accuracy and loss rate. Eventually, the results indicate that the used model compared to other similar methods, Bi-LSTM (bi-directional LSTM), achieves a more efficient preserving a high accuracy level.

show abstract

“…FSN), that is typically applied under the off-line setting. Instead, we applied the Weighted Moving Average (WMA) technique, that uses the content of the current context window to update normalization statistics on-the-fly, weighted by previous context from past windows with an α parameter(Jorge, Giménez, Silvestre-Cerdà, et al 2022). Finally, as Transformer LMs have the inherent capacity of attending to potentially infinite word sequences, history is limited to a given maximum number of words, in order to meet the strict computational time constraints imposed by the streaming scenario.…”

mentioning

confidence: 99%

Streaming Automatic Speech Recognition with Hybrid Architectures and Deep Neural Network Models

Cano¹

View full text Add to dashboard Cite

Current state-of-the-art models based on Long Short-Term Memory (LSTM) networks have been extensively used in ASR to improve performance. However, using LSTMs under a streaming setup is not straightforward due to real-time constraints. In this paper we present a novel streaming decoder that includes a bidirectional LSTM acoustic model as well as an unidirectional LSTM language model to perform the decoding efficiently while keeping the performance comparable to that of an off-line setup. We perform a one-pass decoding using a sliding window scheme for a bidirectional LSTM acoustic model and an LSTM language model. This has been implemented and assessed under a pure streaming setup, and deployed into our production systems. We report WER and latency figures for the well-known Lib-riSpeech and TED-LIUM tasks, obtaining competitive WER results with low-latency responses.

show abstract

Live Streaming Speech Recognition Using Deep Bidirectional LSTM Acoustic Models and Interpolated Language Models

Cited by 15 publications

References 46 publications

comprehensive study based on societal question and answer system using natural language processing with topic modelling for job assistantship in the world industry

comprehensive study based on societal question and answer system using natural language processing with topic modelling for job assistantship in the world industry

Apply an optimized NN model to low-dimensional format speech recognition and exploring the performance with restricted factors

Streaming Automatic Speech Recognition with Hybrid Architectures and Deep Neural Network Models

Contact Info

Product

Resources

About