Multilingual Modulation by Neural Language Codes

Channel:
Microsoft Research
Subscribers:
344,000
Published on ● Video Link: https://www.youtube.com/watch?v=0TQJoghGNDg


Duration: 48:03
1,044 views
10

Multilingual Speech Recognition is a very costly AI problem, as each language and even different accents require their own acoustic models to obtain optimal recognition performance. Even by using the same phone symbols across languages, each language and even accents impose their own colorings or "twangs", a shift in the acoustic realization of sounds. In this talk, I will outline an approach that uses a large multilingual neural network that is modulated by language codes. These codes are generated by an ancillary network that learns to code useful differences between the "twangs" or human languages. This network architecture allows the quick adaptation to languages.

See more at https://www.microsoft.com/en-us/research/video/multilingual-modulation-by-neural-language-codes/



Other Videos By Microsoft Research


2018-10-18Blind Room Parameter Estimation in Real Time from Single-Channel Audio Signals in Noisy Conditions
2018-10-17Accelerating AI with Project Brainwave and Intel FPGAs
2018-10-17Silent Voice: Unnoticeable Voice Input by Ingressive Speech (Full Version)
2018-10-161/3: Rochester Institute of Technology uses Translator to break communication barriers on campus
2018-10-162/3: Rochester Institute of Technology uses Translator to break communication barriers on campus
2018-10-163/3: Rochester Institute of Technology uses Translator to break communication barriers on campus
2018-10-10Leading labs with Dr. Jennifer Chayes
2018-10-09Multiparty Computation Research
2018-10-09Keynote: Inside Microsoft Azure Datacenter Architecture
2018-10-09Sensor Fusion for Learning-based Motion Estimation in VR
2018-10-09Multilingual Modulation by Neural Language Codes
2018-10-03Soundscape Kayaking Scavenger Hunt [Audio Description]
2018-10-03Soundscape Kayaking Scavenger Hunt
2018-10-03The Future is Fusion with Asta Roseway
2018-10-01Fireside Chat with Dario Amodei
2018-10-01Computational Modelling of Human Epilepsy: from Single Neurons to Pathology
2018-09-27Towards a Unified Bayesian Model for Cyber Security
2018-09-27Manifold Learning Yields Insight into Complex Biological State Space
2018-09-27Extending F* in F*: Proof automation and Metaprogramming for Typeclasses
2018-09-26Rochester Institute for Technology uses Microsoft Translator to bridge communication gaps
2018-09-26Chinook Middle School engages with its community using Microsoft Translator


Tags:
microsoft research