Transcend Language Barriers with Custom Machine Learning APIs: um...
summary
A working web app called "multimédia translator" converts YouTube audio to timed SRT subtitles, translates between English, Spanish, German and French, and can optionally generate dubbed audio while preserving silences. It uses off‑the‑shelf ASR, NMT and TTS models from Hugging Face exported to ONNX, served on Triton inference server or OVH endpoints, with GPUs (V100/H100/L4) and S3 model storage for inference; exporting, preprocessing/postprocessing and timestamp mapping are implemented and a live demo shows usable subtitles and dubbed output with acceptable latency.
AI-written summary · about
chapters (17)
- 0:00AI and APIs introduction
- 1:02Team and goals
- 2:25Multimedia and accessibility
- 4:11User needs definition
- 5:14AI model selection
- 8:00Target languages and tasks
- 9:24Model deployment in production
- 12:54Open Source models
- 15:23Model testing and export
- 17:52Deploying with Triton Inference Server
- 23:52API endpoints solutions
- 28:08SRT subtitle generation
- 33:04Preserving silences in translation
- 34:52Voice dubbing with TTS
- 38:20Web app development
- 39:24App demonstration
- 43:32Conclusion
related talks
43mBreaking the language barrier: how do we quickly add multilanguage support in our AI application?
31mShare information by removing language barriers with Amazon Translate
33mMaximize lifetime value of media content
20mMeta's AI Translations Inference
29mUsing Amazon Transcribe to make content searchable and accessible
14mMachine Learning for developers in a hurry
01h 00mBoost your App with Machine Learning APIs
01h 04mAutomating Podcast Localization via OpenAI APIs