AI for Subtitling - Limecraft presentation the 2022 Open Forum
0 likes201 views
Limecraft's platform utilizes AI to streamline the subtitling process for video teams, reducing the time required to subtitle an hour of content from 8 hours to just 4 hours. The document discusses the challenges and solutions in video content creation, emphasizing the importance of integrating human and artificial intelligence to improve efficiency and creativity in workflows. Limecraft's automation tools are designed to enhance productivity while ensuring data privacy and security.
AI for Subtitling - Limecraft presentation the 2022 Open Forum
- 1. Open Forum 10 February 2022 The Workspace for Video Teams 1 Using AI for Subtitling What works and what doesn’t (Contact: info@limecraft.com)
- 2. From raw footage to finished programme Limecraft - The Workspace for Video Teams Technology to improve the quality of work, not to make professionals redundant 2
- 3. IMAGE FROM Associated Press That is up to 150,000 edited items every year or hundreds of thousands of minutes of content – which must be manually transcribed and shot-listed. The scale is vast and therefore the potential to save time is huge. By solving a problem for AP, Limecraft solved a problem for the industry as a whole. Sandy Mc Intyre VP News - Associated Press The creation and distribution of compelling video content takes on increasingly unmanageable proportions 3 More info: www.limecraft.com
- 4. Volume* & Speed Originality Remote & Global Personalisation Video Formats The bigger Picture… SOURCE: * Video Streaming Market forecast excluding Corporate & EdTech video ** IABM © 2020 Video market needs Manual work Expensive Monolithic/On-premise Complicated Fragmented How it was Automated 25-75% faster Cloud-native, SaaS Self-learning AI Smooth integrations How it should be Limecraft automates the grunt work, freeing up time for creative storytelling and adding value to our clients’ business Producers fight for relevance as video consumption keeps growing and changing Limecraft’s solution means that we have reduced the time to subtitle one hour of content from 8h to just 4h, doubling the efficiency. The AI driving it ensures the best possible use of human time. Matt Simpson - Head of Technology, Accessibility services UK’s 2nd broadcaster 4
- 5. AI and Subtitling - How did we get there? 5 2013-2015 2016-2017 2018-2019 2020-2021 Subtitling • WER 5% • Configurable Spotting • Subtitle Editor Versioning • Machine translation • Timecode management Transcription • Machine transcription • Transcription Editor • WER 10% Neuro-linguistic Programming • Better word recognition (WER <2%) • Better diarisation • Better spotting Transcription Quality - Efficiency Gain Use Cases • Transcribing Interviews • Archive Indexing Use Cases • Same language subtitling • Closed Captioning Use Cases • Localisation
- 6. The MeMAD Project 6 3,5Mio R&D Horizon 2020 project (2018 – 2021) Smart Tools for handling digital content in creative industries How can you take advantage of AI? – Will does affect your workflow? – How to deal with inaccuracy? – So, which applications do you need on top? Spoiler: take the best of both worlds and combine Artificial and Human Intelligence More info: www.memad.eu
- 7. The MeMAD Project 7 Methodology: “Agile Development” • Data cleaning and training • Iterative Prototyping • Subjective Testing More info: www.memad.eu
- 8. First iteration: Fast but not pleasant Differences between language combinations (or between the translators…) Tedious, annoying, limiting… Spotting and segmentation was poor
- 9. Efficiency = rewiring the workflow ● Subtitle template + MT vs. ASR + MT ● Garbage in = More Garbage out ● QC the transcript before spotting!!! Tangible improvements in automatic spotting Second iteration: Focus on the process
- 10. Methodology : tuning and testing 10 More info: https://www.limecraft.com/cases/itv/
- 11. LIMECRAFT connected creativity How can you take advantage of AI? – How will it affect your workflow? – How to deal with inaccuracy? – Which applications do you need on top? Lessons Learned - Never leave AI unattended – Repeated error correction is a showstopper. When using ASR, Word Error Rates must be lower than 2% – Fine-tuning is a step-by-step and iterative process – Usability is key - there is no one size fits all UI Best practices – ASR: relentlesly look for the best, not the cheapest. Look for autonomous training, custom dictionaries,… – Spotting: look for configurable spotting rules – Post-editing: look for solutions easily fits your workflow
- 12. Benchmark ASR solutions : look for the best 12
- 13. Custom Dictionaries : be as specific as possible to avoid falst positives 13
- 14. Spotting - configurable rules and multiple settings 14
- 15. LIMECRAFT connected creativity Copyright 2022 Limecraft , NV. Confidential – Distribution prohibited without permission Limecraft ® Flow for Subtitling Computer assisted Subtitling to save up to 75% of your time AI-powered ________________________ ____ Limecraft instantly turns audio into timed text with the highest possible accuracy, translates scripts, and automatically creates properly styled and timed subtitles. The Workspace _______________________ _____ A single Workspace for language professionals allows them to upload media, create and QC transcripts and subtitles, and to dispatch their work. Ease of Integration _________________________ ___ Using a library of plug-ins and adapters, Limecraft easily integrates as part of your operating processes and current application landscape. Committed to Security _______________________ _____ Maintain total control and privacy over your data. No content is kept by us, or shared with third parties.
- 16. LIMECRAFT connected creativity Copyright 2022 Limecraft , NV. Confidential – Distribution prohibited without permission
- 17. Limecraft offers professionals a collaborative workspace for efficiently creating and distributing the most engaging stories IMAGE FROM Undercover - Newen More info and free try : www.limecraft.com 17 (Contact: info@limecraft.com)