Fine-tuned Whisper models for five languages, built for subtitle and indexing workflows.
- aBuilt and aligned over 1,200 hours of domain-specific entertainment speech with timestamp-level quality controls.
- bFine-tuned Whisper small and medium with targeted LoRA adapters for code-switching and low-latency scenarios.
- cDelivered sub-15% WER in noisy media conditions, unlocking scalable subtitle and indexing workflows.
