CaptionSentry Q&A

Last week we had a fantastic turnout for our free webinar "CaptionSentry: An In-Depth Look into the Premier Real-Time Automated Captioning Solution from NCI." At the end, we had time for participant questions and we thought we’d share the answers with you below.

If you’ve been looking for an Automated Captioning Solution for your company, please reach out and we will be happy to guide you through the process, answer any additional questions you might have, and set you up for a live demo. We believe CaptionSentry is the best ASR option on the market and we love chatting about it!

Question: Is this service only for live streaming/broadcasting or can it be used for pre-recorded projects?

Answer: NCI has been working on an offline tool, but our development focus to date has been on our real-time ASR tool.

Question: What are your plans for additional languages? Amharic, French, Mandarin?  Timing?

Answer: The only languages we currently support are English and Spanish but we can consider offering/supporting additional languages should there be a compelling demand.  Also, we have a solution for auto translation from our English or Spanish captions into more than 40 languages. For more information please go to https://www.ncicap.org/live-captioning-translation

Question: Do you mix ASR only and human captioners in the same streams? or is it an either or option?

Answer: It is not an either/or solution. Humans can take over from CaptionSentry, and CaptionSentry can take over from humans such that the data from either source can be mixed in the same stream. Additionally, we have the failsafe feature that can be activated to allow for CaptionSentry to send captions if the human captioner loses their connection and stops sending data.

Question: You mentioned that you deliver captions for broadcast and meetings; are there other applications you can mention? 

Answer: CaptionSentry is currently being utilized by broadcast clients and it can also work with online meetings/webinars such as the Zoom platform.

Question: Does your Cognitive Accuracy have to be vetted by the FCC in any capacity? Will you ask them to evaluate its accuracy?

Answer: The Cognitive accuracy has not been vetted by the FCC, and it is not something NCI is holding out to be a standard for the industry. Cognitive accuracy is a measurement standard that was devised by NCI.  It also uses audio comparison. It is based on the ability of the viewer to comprehend the intended meaning of the captions. Errors are assessed only if the intended meaning was not able to be understood. Missing or incorrect punctuation and missing or incorrect speaker changes are not counted as errors. We don’t know if we will ask the FCC to evaluate. 

Question: Does it do speaker Identification?

Answer: Although CaptionSentry has the ability to produce speaker tags (chevrons), we’ve only been able to get them to appear correctly, on average, 50% of the time and there are unfortunately a lot of false positives (or incorrectly added speaker changes) as well. We have opted not to put these on the air until we can achieve a much higher rate of speaker change accuracy, unless specifically requested to do so.

We continue to analyze features, industry developments, and competitive products to ensure we are offering better ASR captioning than anyone else. In our testing of other ASR products, we have found that none offer accurate speaker changes more than 50% of the time, at least not consistently.

Question: Have you worked with the FCC on this solution?

Answer: We have kept them informed of developments in ASR including its improved quality, its applications, and its impact on the industry. We have also met with them in person in past years to specifically review how ASR impacts various stakeholders.

Question: Are there any facilities for training the recognition system for proper nouns prior to launching the service for our programs?

Answer: In order to help increase CaptionSentry’s accuracy we have an available team of Automated Captioning Specialists who enter and maintain a word list.

Question: What is the average latency of Caption Sentry?

Answer: Latency can be as low as 2 seconds if an NCIStream device is place onsite.  If the client prefers not to have a device onsite, latency is still only 3-4 seconds.

Question: Does the latency increase with fast talkers?

Answer: Latency does not increase with fast talking.  There will be a few instances where the system will appear to stall for a few seconds, adding latency for that particular moment, but on average it is about a 2-second response time if an NCIStream is onsite.