Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Apr 23, 2026

Author Spotlight: Investigating the Impact of Emotional Prosodies on Voice Recognition and Perception
05:48

Author Spotlight: Investigating the Impact of Emotional Prosodies on Voice Recognition and Perception

Published on: August 9, 2024

1.8K

Frame-by-frame language identification in short utterances using deep neural networks.

Javier Gonzalez-Dominguez1, Ignacio Lopez-Moreno2, Pedro J Moreno2

  • 1Google Inc., NY, USA; ATVS-Biometric Recognition Group, Universidad Autonoma de Madrid, Madrid, Spain.

Neural Networks : the Official Journal of the International Neural Network Society
|September 23, 2014
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Metabolomic profiling in phenylketonuria: a systematic review of human studies.

Metabolomics : Official journal of the Metabolomic Society·2026
Same author

A response to EA-4/23 INF:2025 "The Assessment and Accreditation of Opinions and Interpretations using ISO/IEC 17025:2017".

Forensic science international·2025
Same author

Diffusivity alterations related to cognitive performance and phenylalanine levels in early-treated adults with phenylketonuria.

Journal of neurodevelopmental disorders·2025
Same author

Volumetric brain reductions in adult patients with phenylketonuria and their relationship with blood phenylalanine levels.

Journal of neurodevelopmental disorders·2024
Same author

Impact of supplementation with dihydroxylated vitamin D<sub>3</sub> on performance parameters and gut health in weaned Iberian piglets under indoor/outdoor conditions.

Porcine health management·2023
Same author

Validations of an alpha version of the E<sup>3</sup> Forensic Speech Science System (E<sup>3</sup>FS<sup>3</sup>) core software tools.

Forensic science international. Synergy·2022

Deep neural networks (DNNs) significantly improve automatic language identification (LID) for short speech utterances. This approach offers real-time performance, outperforming traditional i-vector systems by up to 76%.

Area of Science:

  • Speech processing
  • Machine learning
  • Computational linguistics

Background:

  • Deep neural networks (DNNs) have shown success in acoustic modeling for speech recognition.
  • Automatic language identification (LID) is crucial for various real-time applications.
  • Short test utterances pose challenges for traditional LID systems.

Purpose of the Study:

  • To adapt DNNs for automatic language identification (LID) using short acoustic features.
  • To evaluate the real-time capabilities of DNN-based LID systems.
  • To analyze factors influencing DNN performance in LID.

Main Methods:

  • Utilizing deep neural networks (DNNs) to process short-term acoustic features for language identification.
  • Analyzing system performance based on training data size, network architecture, context, and utterance duration.
Keywords:
DNNsReal-time LIDi-vectors

Related Experiment Videos

Last Updated: Apr 23, 2026

Author Spotlight: Investigating the Impact of Emotional Prosodies on Voice Recognition and Perception
05:48

Author Spotlight: Investigating the Impact of Emotional Prosodies on Voice Recognition and Perception

Published on: August 9, 2024

1.8K
  • Developing methods for combining frame-by-frame language identification posteriors.
  • Conducting experiments on the NIST LRE09 (3s task) and a large-scale Google 5M LID dataset.

    • Main Results:

    • DNNs demonstrate superior performance compared to i-vector systems for LID on short utterances.
    • Relative improvements of 40% on the LRE09 3-second task and 76% on the Google 5M LID dataset were achieved.
    • DNNs are suitable for real-time LID applications due to their frame-by-frame posterior emission capability.

    Conclusions:

    • Deep neural networks are highly effective for automatic language identification, especially for short utterances.
    • The proposed DNN-based approach offers significant performance gains and real-time applicability.
    • Further research can explore optimal system configurations and advanced posterior combination techniques.