Live Translate · published
Live Translate uses the full power of your computer
Live Translate runs entirely on your own computer. Translation, subtitle processing, and AI tasks are performed locally whenever possible, keeping your data private and under your control.
Because of this, performance depends on your hardware. A faster CPU and GPU provide quicker translations, shorter subtitle processing times, and a smoother overall experience. More modest systems may require additional processing time.
We are continuously improving Live Translate. In the near future, we will introduce more advanced AI models, higher performance, and better quality through regular updates. Our goal is to keep Live Translate up to date while making the best possible use of your existing hardware.
Subtitle burning is now significantly faster
On many systems, the GPU vendor's native video encoder (NVIDIA NVENC, AMD AMF, or Intel Quick Sync) could not be initialized, causing subtitle burning to silently fall back to the CPU. As a result, a process that was expected to be hardware-accelerated was actually running on the processor.
This release introduces the Windows Media Foundation encoder. It works independently of the GPU vendor and can utilize the hardware acceleration available on NVIDIA, AMD, and Intel graphics hardware. If the vendor's native encoder is available, it is still preferred. If not, the process now falls back to Windows Media Foundation instead of the CPU, allowing hardware acceleration to remain enabled.
Benchmark (20-second 1080p video)
Before: 6.3 seconds, 26.5 MB
After: 2.2 seconds, 6.1 MB
Results:
Approximately 2.9x faster.
Output file size reduced to approximately one quarter.
We also fixed the encoder self-test logic. Previously, the application validated the encoder using simplified settings but performed the actual subtitle burn with different quality parameters. This could cause an encoder that passed the test to fail during real processing.
The validation process now uses exactly the same settings as the actual subtitle burn, ensuring that an encoder that passes the test will also complete the job successfully.