Hey all, Koemei founder here. We developed Koemei from a simple question: everyone agrees that video is the new text. In his interview, Facebook founder Mark Zuckerberg said "in under 5 years, more than 90% of Facebook will be video..". The same goes in the enterprises, the use of is video is now replacing text in learning, product marketing, sales etc. But why can't we search videos just like we search text?
I decided to build Koemei with my co-founders after working with +30 people to look through nearly 40,000 videos searching for specific information. If there was a product like Koemei, this experience would have cost 90% less in money, manpower and time.
Report
Great idea but solution is not perfect, far from it. I tried with an interview and transcript isn't good. Otherwise, i thing lot of company needs transcription solution. there are a big market !
@fabs_31 Machine transcription is not perfect yet, but great progress has been achieved in the last couple few years alone. The question to ask is what can we do with an imperfect machine transcription? A lot! Thanks to machine learning and natural language processing, we can in fact use imperfect transcript to extract nearly all valuable insights, topics, keywords, entities etc. from a video at a significantly higher accuracy than ever before. This is phenomenal. Our goal at Koemei is to keep leading this pack with the best machine transcription technology in the world and easy to use video search. Thanks for your comment.
@fabs_31 Thanks!
As Temi explained, at Koemei, we are overcoming the complexity of the automatic speech recognition problem by coupling it with deep learning and natural language processing. We're positioning our product for enterprise video such as training videos, sales pitch videos, meetings, and conferences. The three technologies combines will enable employees to capture topics of interest inside work-related videos.
However, there are some tips which we suggest for you to attain a great degree of accuracy:
1- Upload a video that is > 3 mins long.
2- Choose a video that has no music in the background.
3- Choose a video with native English speakers to obtain high accuracy ( >90%).
Im happy to answer your further questions.
Report
@wehbewalaa@fabs_31 Thanks for the reply ! I know importance of "model" in your case. More data you have, more efficient you are.
@fabs_31 Exactly! We are also studying "teaching" our system "common video" so that we can better extract transcripts.
We have some exciting projects in the pipeline that will add more value to the product and will allow transcription of noisy videos.
Hey, it's Walaa from Koemei.
Koemei has speech-to-text APIs in addition to a video search widget, which you can embed on any website. This makes Koemei technology at your fingertips. I'd like to share a demo we've worked on with the World Economic Forum where Koemei's video search widget was embedded on their website. Here's how it looks like: http://demo.koemei.com/wef_video.... I'd love to hear from those who are interested in having video search integrated on their website ;)
Report
Great product! How is it different from Panopto or YuJa for exemple?
Koemei
Castle
