At-Scale OCR Of Television News Experiments: Optimizing The Still Frame File Storage Format
Analyzing petascale video archives poses unique computational challenges, from the underlying processor and accelerator requirements to simply moving that much…
From LSM's To LMMs For ASR: Evaluating Gemini's Performance At Transcribing An Evening News Broadcast
As we continue to evaluate the rapid progress of large model ASR systems, from lightly to heavily generative LSMs to…
Comparing GCP's Chirp & Chirp 2 ASR Models: Dropping Entire Passages
Yesterday we examined how GCP's new Chirp 2 ASR model hallucinates speech during non-verbal musical interludes in news broadcasts, resulting…
Comparing GCP's Chirp & Chirp 2 ASR Models: Hallucinating Speech During Music
Over the past six months we have continued to compare GCP's Chirp and Chirp 2 ASR models, each time finding…
Audience-Specific Podcasts: Customizing Our Daily "Top Stories" Biosurveillance Podcast Concept For Experts, Policymakers & The American Public
Yesterday we demonstrated feeding a daily roundup of global disease outbreak news headlines from around the world into a "thinking"…
A Daily "Top Stories" Global Disease Outbreak Podcast Concept Using GCP's Gemini 2.0 Thinking + Text-to-Speech API
What might it look like to feed a daily roundup of global disease outbreak news headlines in all the world's…
Using GCP's Chirp + Gemini 1.5 Pro + Speech-To-Text API To Summarize A Day Of Russian TV News Into A 3 Minute "Top Stories" Podcast
What might it look like to use GCP's Speech-to-Text API's Chirp LSM model to machine transcribe a full day of…
A Daily "Top Stories" Global Investment News Podcast Concept Using GCP's Gemini 2.0 Thinking + Text-to-Speech API
What might it look like to feed a daily roundup of global investment news headlines in all the world's languages…
A Daily "Top Stories About NVIDIA" News Podcast Concept Using GCP's Gemini 2.0 Thinking + Text-to-Speech API
What might it look like to feed a daily roundup of news headlines about NVIDIA from across the world in…
At-Scale OCR Of Television News Experiments: OCR'ing 10 Billion Seconds Of Global TV News For Just $47.5K Vs $26.9M
In collaboration with the Internet Archive's Television News Archive, we have successfully OCR'd 4.2 million television news broadcasts from around…
Behind The Scenes: Identifying Failed Recordings: Using Large Multimodal Modals Like ChatGPT & Gemini: Part 3
Continuing our series examining whether Large Multimodal Models (LMMs) like ChatGPT and Gemini might be able to help us identify…
Behind The Scenes: Identifying Failed Recordings: Using Large Multimodal Modals Like ChatGPT & Gemini: Part 2
Earlier this week we demonstrated the limitations of using Large Multimodal Models (LMMs) like ChatGPT and Gemini to detect corrupted…
Behind The Scenes: Identifying Failed Recordings: Using Large Multimodal Modals Like ChatGPT & Gemini
As we continue our efforts to scan the TV News Archive for failed recordings, how might Large Multimodal Models (LMMs)…
The Influence of Media Propaganda on Green Housing Consumption in China Based on GDELT Big Data
As part of China's two-carbon strategy, green buildings are a vital component in addressing climate change.Formulating a media propaganda strategy to…
Behind The Scenes: Identifying Failed Recordings: Examining Curation Metadata
Any large longitudinal audiovisual archive will have some number of recordings that suffer from technical errors, ranging from minor audio…
Behind The Scenes: API Quotas & The Impact Of A Fraction Of A QPS
All hosted APIs have rate-limited quotas of some form to protect them from abuse and to ensure equal sharing of…
Behind The Scenes: A Look Back At A Month Of Real-World AI API Latency At Scale
Last week we examined a 24-hour period of real-world AI API latency and error rates as an illustration of the…
Behind The Scenes: GCP's Network Intelligence Performance Dashboard
The sheer massiveness of GCP's core service offerings is such that there are a wealth of hidden gems buried within…
Behind The Scenes: GCP Network Intelligence Topology Mapping Of Our OCR Cluster At Startup
GCP's Network Intelligence service offers an incredibly powerful Network Topology visualization that shows all of the various GCP services being…
Behind The Scenes: Using Our Bigtable + BigQuery + GCS Digital Twin To Queue Missing Broadcasts
Our massive new collaboration with the Internet Archive to OCR its complete quarter-century Television News Archive spans ten million broadcasts…
At-Scale OCR Of Television News Experiments: Comparing 1FPS Frame Extraction Quality Levels
One of the most remarkable and difficult challenges of scaling novel workflows from one-off experiments to massive archive-scale production implementations…
At-Scale OCR Of Television News Experiments: Ornamental Vs Chyron Text
One of the great tradeoffs in using image montaging to achieve a 100-200x performance increase and cost reduction for at-scale…
At-Scale OCR Of Television News Experiments: Vertical + Horizonal Text – Results From A Taiwanese Broadcast
Unlike most of the world, television news broadcasts in Taiwan make regular use of vertical onscreen text that appears alongside…
At-Scale OCR Of Television News Experiments: Results From A Sample Circa-2018 South Sudan Broadcast
Below is a transcribed OCR excerpt from a circa-2018 South Sudanese broadcast. As with our previous examples, the difficulties of…