Conversational telephone speech recognition for Lithuanian

Lileikytė, Rasa; Lamel, Lori; Gauvain, Jean‐Luc; Gorin, Arseniy

doi:10.1016/j.csl.2017.11.005

Cited by 10 publications

(2 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There has been work on improving ASR recognition accuracy in agglutinative languages by splitting the words into morphemes. The focus of our experiments is to address OOV recovery and WER reduction for any low-resource language in general and, is complementary to the technique of using morphemes as subwords [65,66,67,68].…”

Section: Oov Problem In Low Resource Agglutinative and Inflective Lan...mentioning

confidence: 99%

Low Resource Kannada Speech Recognition using Lattice Rescoring and Speech Synthesis

Murthy¹,

Sitaram²

2023

IJST

View full text Add to dashboard Cite

Objectives: Improving the accuracy of low resource speech recognition in a model trained on only 4 hours of transcribed continuous speech in Kannada language, using data augmentation. Methods: Baseline language model is augmented with unigram counts of words, that are present in the Wikipedia text corpus but absent in the baseline, for initial decoding. Lattice rescoring is then applied using the language model augmented with Wikipedia text. Speech synthesis-based augmentation with multi-speaker syllable-based synthesis, using voices in Kannada and cross-lingual Telugu languages, is employed. We synthesize basic syllables, syllables with consonant conjuncts, and words that contain syllables that are absent in the training speech, for Kannada language. Findings: An overall word error rate (WER) of 9.04% is achieved over a baseline WER of 40.93%. Language model augmentation and lattice rescoring gives an absolute improvement of 16.68%. Applying our method of syllable-based speech synthesis over language model augmentation and rescoring yields a total reduction of 31.89% in WER. The proposed approach of language model augmentation is memory efficient and consumes only 1/8 th the memory required for decoding with Wikipedia augmented language model (2 gigabytes versus 18 gigabytes) while giving comparable WER (22.95% for Wikipedia versus 24.25% for our method). Augmentation with synthesized syllables enhances the ability of the speech recognition model to recognize basic sounds thus improving recognition of out-of-vocabulary words to 90% and in-vocabulary words to 97%. Novelty: We propose novel methods of language model augmentation and synthesis-based augmentation to achieve low WER for a speech recognition model trained on only 4 hours of continuous speech. Obtaining high recognition accuracy (or low WER) for very small speech corpus is a challenge. In this paper, we demonstrate that high accuracy can be achieved using data augmentation for a small corpus-based speech recognition.

show abstract

Section: Oov Problem In Low Resource Agglutinative and Inflective Lan...mentioning

confidence: 99%

Low Resource Kannada Speech Recognition using Lattice Rescoring and Speech Synthesis

Murthy¹,

Sitaram²

2023

IJST

View full text Add to dashboard Cite

show abstract

“…Our research has only approached the English language. However, it is clear that we must take into consideration variations that can appear in other languages, especially for transcribing speeches [13]. In addition to sentiment analysis, more complex approaches can be used for customer satisfaction, such as those presented in [14].…”

Section: Literature Reviewmentioning

confidence: 99%

Call-Center Virtual Assistant Using Natural Language Processing and Speech Recognition

Vasilățeanu

Ene

2018

J. ICT des. eng. technol. sci.

View full text Add to dashboard Cite

Call center assistance is one of the many domains of activity that could be enabled by Arti icial Intelligence. Enter Customer Recommended Interaction Software (CRIS). The idea of a virtual agent that can offer assistance during a live call or troubleshooting procedure has great potential and can be used to unlock a great extent of advantages. The virtual agent presents a dashboard to the employee, with relevant facts about the conversation presented in real-time. The dashboard contains elements such as call-category, based on detected keywords, and sentiment analysis. We propose a proof of concept that uses state-of-the-art cloud-computing technologies to lay those basic functionalities that we have envisioned as our solution. One of our objectives is to test available NLP and cloud platforms. We follow a case-study approach to the research, in which the prototype is tested in different scenarios, including short and long conversations, and the results are examined and discussed. In more detail, we involve nine speakers, with different accents and speaking styles, in 85 scripts and 3 real-life conversations. Results are promising, with good accuracies for short conversations and limitations due to the lack of domain-speci ic knowledge bases applicable to the call-center work.

show abstract

Automatic Speech Recognition System for Tonal Languages: State-of-the-Art Survey

Kaur

Singh

Kadyan

2020

Arch Computat Methods Eng

View full text Add to dashboard Cite

Conversational telephone speech recognition for Lithuanian

Cited by 10 publications

References 6 publications

Low Resource Kannada Speech Recognition using Lattice Rescoring and Speech Synthesis

Low Resource Kannada Speech Recognition using Lattice Rescoring and Speech Synthesis

Call-Center Virtual Assistant Using Natural Language Processing and Speech Recognition

Automatic Speech Recognition System for Tonal Languages: State-of-the-Art Survey

Contact Info

Product

Resources

About