Cobalt Labs

What are the cool things that we are working on

Cobalt Call
Summarization TM

Topics, Questions, Action Items and Follow-Ups

Call center agents often need to write an overview after each call, describing the main issues and how they were resolved. Cobalt’s Call Summarization project leverages multiple machine learning techniques:

  1. To identify the most relevant lines from the call transcript, tagging them as topics, customer questions, agent actions, and follow-ups
  2. To draft a human-friendly paragraph summarizing the entire call

The agent can greatly reduce their post-call processing time by accepting or lightly editing the suggested summary rather than having to write it from scratch.

Acoustic Events TM

Gaining Intelligence Through Sound and Acoustic Events

In addition to speech, Cobalt can train models to recognize other sounds.  One lab project built a model to recognize sounds that could be useful for identifying security incidents, such as breaking glass, screaming, alarms, etc. 

Imagine a Public Security System of detecting dreams, thuds, crash sounds, etc.

Under Resourced Languages & Non-Profit Support TM

Cobalt Speech Scientists and Researchers donate their time and expertise to help non-profit groups build speech technology for languages that are severely overlooked.

For Example:

  1. Bilaspuri, Dogri, and Haryanvi, spoken in India (see research results at
  2. Eastern Min, spoken in southeast China and parts of Taiwan
  3. Khmer, spoken in Cambodian (in support of the Cambodian Oral History Project)