Rev AI Q1 2022 Product Update
With 2022 well underway, we wanted to share some exciting highlights from the last few months for Rev AI and our users. We’ve released some exciting new functionality and made Rev AI more accessible to developers – and there’s much more to come! Here are some of our recent developments:
We built an entirely new “v2” ASR engine, now in Open Beta and heading to General Availability on March 7, 2022
Our new model, which we’re calling Rev AI’s v2 ASR model, outperforms our existing production (“v1”) model (and all the competition from what we can see!) by a significant margin. (See table below)
We performed a comparison of our new model against Google, Amazon, Microsoft, and others, and were quite proud with the results:
We made our cutting-edge features available in Beta
We launched Rev AI Labs, a collection of our latest features for initial Beta use. Three new features are now available:
- Rev AI Human Transcription – now you can use Rev AI to request both machine and human generated transcripts and captions. This can be done at the whole file level or for just parts of a file with timestamps preserved so you can easily combine ASR and human outputs. This unlocks new levels of accuracy while controlling costs for you and your customers.
- Topic Extraction – you can now extract topics with scores and corroborating statements to better understand your content. This is great for auto-tagging and it also pairs nicely with the Rev AI Human Transcription capability above (e.g., to get human quality on the topics you care most about).
- Sentiment Analysis – you can detect sentiments like “positive”, “neutral”, or “negative” with scores and corroborating statements to better understand your content. Like Topic Extraction, this also pairs nicely with the Rev AI Human Transcription capability above.
Interested in trying out these new features? You can – they’re available to all our customers as part of our Open Beta program!
We’re the leader (and continuing to invest) in reducing bias in ASR
Our production and v2 engines were independently benchmarked by our customer, HireVue, against other leading ASR providers on dimensions of gender, ethnic, and country-of-origin/accent. You can learn more in this talk that Joel Susal, Rev AI Head of Product, gave at the AI Summit in NY in December. (spoiler: our production model outperforms Google & Amazon and our v2 model categorically outperforms our v1 production model).
We contributed useful tools and data to the community
We released our first open source project, fstalign, which aligns a transcript with its source audio for easy and accurate Word Error Rate (WER) comparison. We use this in our benchmarking tools and now you can too.
We also released our first public dataset, earnings21, which you can download and use in your own WER testing to evaluate ASR engines. For further reading, check out our publications on fstalign here and earnings 21 here.
Dan Kokotov, Rev co-founder and VP of Engineering, shared predictions on the future of ASR at the AI Summit
Dan shared his predictions on multilingual models, ASR at scale, and responsible AI among other topics in his presentation.
We made it easier to explore our APIs with Postman
With our Postman collection, you easily test and interact directly with the Rev AI API without having to write any code. Postman is a free to use development tool for working with APIs. Check out our documentation or access the Rev AI Postman collection directly.
We made it easier to add ASR to video applications and workflows
We integrated with AV Flow, the leading low/no code video platform. With this integration, all AV Flow developers can now use Rev AI’s ASR for key video features like transcription, captioning, and subtitling.
We improved our security
Rev AI is now SOC 2 Type II Security Compliance Certified. Keeping your data safe is one of our highest priorities here at Rev. This certification verifies that our systems are protected against unauthorized access, yet are available for operation and use by you, our customers.
We added new async and streaming languages in beta
We continue to grow our global language support for our Rev AI customers. We have five more languages in our async service: Indonesian, Hebrew, Tamil, Telugu and Farsi. For our streaming offering customers can now transcribe their files in Mandarin, German, Spanish, French, Italian, Japanese, Korean, Portuguese.
We enhanced usage and jobs dashboards
We’re revamping our Usage and Jobs Dashboards to reflect new jobs and features available from Rev AI. You’ll see summaries of the number of jobs, hours, and cost of your ASR Sentiment Analysis, Rev AI Human Transcription, and Topic Extraction jobs in the updated dashboard.
We now offer language ID in closed beta
Our API can now help predict the likelihood of a language being dominant in a file. It is currently built only for our async service and supports 22 languages.
And we have more on the way…
We are investing heavily in making Rev AI the best speech platform in the world. More features, improvements, and updates coming throughout 2022. Stay tuned!