Elastic Silicon Valley Meetup - Using Elastic to power voice search and analytics at VoiceBase

Using Elastic to power voice search and analytics at VoiceBase

At VoiceBase, we use Elastic to power search and analytics based on recorded speech, driven by our speech-to-text, semantic indexing, and predictive technologies. In addition, Elastic is helping our customers solve key problems of multi-tenancy, rich metadata integration, and interactive phrase ("keyword") spotting. In this talk, we will share how we integrate Elastic into our offering, and practical lessons learned along the way.

John Bocharov (Director of Engineering @ VoiceBase) manages a team of high-caliber engineers responsible for orchestrating speech analytics, over thousands of compute cores. Previously, he and his teams built a recommendation engine with Lambda architecture (before it was cool) and products for adaptive bit rate streaming over HTTP (way before it was cool). John is passionate about data products, fútbol, and toe socks. In fact, he is probably wearing a pair right now.

Transcript

1. www.voicebase.com Indexing Voice Data for Fun and Profit John A. Bocharov Director, Engineering john@voicebase.com
2. www.voicebase.com Agenda •  Who? •  Why? •  What? •  How? •  Huh?!? John A. Bocharov | john@voicebase.com
3. www.voicebase.com … in 90 seconds or less John A. Bocharov | john@voicebase.com
4. www.voicebase.com Use Cases •  Transcripts •  Keywords •  Predictions •  Search •  Analytics John A. Bocharov | john@voicebase.com
5. www.voicebase.com Search •  /media?query=Zaphod John A. Bocharov | john@voicebase.com
6. www.voicebase.com Pagination (Cursor-Based) •  /media?query=Zaphod&after=d85bab31&limit=25 John A. Bocharov | john@voicebase.com
7. www.voicebase.com Keyword Spotting •  Key Words or Phrases (I know what to look for) •  At upload time (production) •  Interactively (exploration) –  /media?query=Dent,Trillian,Marvin,Prefect John A. Bocharov | john@voicebase.com
8. www.voicebase.com Architecture (Excerpted) John A. Bocharov | john@voicebase.com VoiceBase API Speech Indexing Analy6cs Elas6c Upload Query
9. www.voicebase.com Metadata •  { “metadata” : { “extended” : { “agent” : “Smith” } } } •  /definitions/media/search ! agent •  /media ?filter.metadata.extended.agent=Smith &query=Mr.+Anderson John A. Bocharov | john@voicebase.com
10. www.voicebase.com Multi-Tenancy •  { “metadata” : { “extended” : { “tenant” : “Meta Cortex” } } } •  /definitions/media/search ! tenant •  /media ?filter.metadata.extended.tenant=Meta+Cortex &query=Neo •  /media&filter.metadata.extended.tenant=Meta+Cortex &filter.metadata.extended.agent=Smith &query=Neo John A. Bocharov | john@voicebase.com
11. www.voicebase.com Practical Lessons •  ELB •  Socket vs REST •  Monitor, Monitor, Monitor –  Synthetic monitoring (3rd party + in-house) John A. Bocharov | john@voicebase.com
12. www.voicebase.com Thought of the Day “Oh Goodie! Yet another query to optimize.” – Every DBA, Ever John A. Bocharov | john@voicebase.com
13. www.voicebase.com Scalability •  Elastic, then Database –  /media?query=Neo&filter.created.gte=2016-08-22 •  Straight Database Queries –  /media?after=d85bab31&limit=25 –  /media?filter.created.lte=2016-08-22 –  /media?filter.created.lte=2016-08-22&after=d85bab31&limit=25 •  Elastic, then Database –  All the Things John A. Bocharov | john@voicebase.com
14. www.voicebase.com An Unanswerable Question… Maybe •  All your BASE are belong to us. •  ALL your base are belong to us. •  All your base are belong TO US. John A. Bocharov | john@voicebase.com
15. www.voicebase.com … Q & A John A. Bocharov | john@voicebase.com bit.ly/2bgAsNN