improving analytics with machine and deep learning - jeh daruvala - speechtek - 10 aug 2015

15
Improving Analytics with Machine and Deep Learning SpeechTEK 2015 Jeh Daruvala - CEO [email protected]

Upload: jeh-daruvala

Post on 11-Feb-2017

360 views

Category:

Documents


2 download

TRANSCRIPT

PowerPoint Presentation

Improving Analytics with Machine and Deep Learning

SpeechTEK 2015

Jeh Daruvala - [email protected]

1

Why Speech Analytics?

Voice-of-the-Customer

Compliance

Cost Savings (Automated agent monitoring)

Audio Search

2

OmniTraq - Omni-channel Business Intelligence3ComprehensiveIntelligenceAnalytics

Media MonitoringCall Center SolutionsCoreTraq

Actionable Insights

Local B2B lead generationDevelop competitive engagement strategiesOmni-channel Voice-of-the-Customer discoveryBrand sentiment discoveryEnsure regulatory & compliance adherenceEvaluate & manage performance of call centersTurn internal documents, calls, meetings & events into searchable metadataCustomize vocabularies 10-100X Faster, Better, Cheaper

Custom VocabulariesAccurate B2B Speech Systems need to understand,Product and Brand namesIndustry and Company specific Business Terms

Mass customized vocabularies (Unique to each customer)

Speech based Semantics ( Speech + Text Analytics)

Language Model (LM) based SystemsLegacy Phonetic Vocabularies < 1000 wordsLanguage Models can contain tens of millions of sentencesYet LM based systems can save $ millions when building Custom Vocabularies !4

Yactraqs Core Technology5CoreTraq Platform

Self-learning capability allows for perpetual and inexpensive enhancements to the platformSpeechRecognizerText Transcripts

Machine Learning(Offline)

Provides ongoing improvements to Speech RecognizerNLU*Topic Engine

*Natural LanguageUnderstanding

Audio & Video Inputs

Phone CallsRadioTVWeb VideosPodcastsMore

Output Metadata

KeywordsTopicsTime-taggingSentimentWord confidence

Audio & video is not yet natively understood by computers, which work with character data.CoreTraq is Yactraqs unique speech based semantic platform that accurately converts audio & video into character data.

5

6Example Video - Wells Fargo Becomes The Most Valuable Bank In The World: Bottom Line | CNBC

7Wells Fargo Becomes The Most Valuable Bank In The World: Bottom Line | CNBChttps://www.youtube.com/watch?v=J4OMX1Gq5ykGeneric vocabulary fails to classify video

Wells Fargo Becomes The Most Valuable Bank In The World: Bottom Line | CNBChttps://www.youtube.com/watch?v=J4OMX1Gq5yk8

Financial Services vocabulary offers dramatic improvement 8.52 score is among the highest confidence numbers in Yactraq Speech2Topics

AppendixImproving Analytics with Machine and Deep Learning

SpeechTEK 2015

Jeh Daruvala - [email protected]

9

Complex Language ModelingWeb Crawler based Linguistic DataGeneral language awarenessGeneric vertical vocabulariesFinancial, Retail, Healthcare etcIs A similar to B? type questions General call/audio classification

Class Based Language ModelsLimited linguistic data (Phone call transcripts)What specifically? type questionsWhy 5 calls and not 1 to resolve this?10

IP StrategyPatent pending Speech2Topics Technology

IP for Defense partnership with

High degree of Freedom-To-Operate

Access to Intellectual Ventures patent portfolio of over 40,000 patents11

Dynamic Language ModelsConfiguration API

Continuous Configuration FeedTarget Topics and Entities

Automated Vocabulary Updates

12

Versatile Speech RecognitionMachine learning based information compressionHigher accuracyEmbedded systems

LVCSR Well suited to small footprint speech recognizers

Deep Neural Networks For acoustically challenging dataAccentsLanguages13

Hospitals Wary of Hackers Seek Insurance from AIGhttps://www.youtube.com/watch?v=Dm2O8XyEmo0

14

General vocabulary shows a high level understanding of topics

Hospitals Wary of Hackers Seek Insurance from AIGhttps://www.youtube.com/watch?v=Dm2O8XyEmo0

15

Financial Services vocabulary has sharper understanding and picks up specific reference to Blue Cross