Live Speech needs to be simultaneously translated into multiple target languages.

Source summary

Develop live translation of speeches into Indian languages using BHASHINI models and APIs, with captions and broadcast-compatible outputs. The source calls for an open-source, cloud-based, platform-independent service, multi-user scale, a management dashboard and mobile or web access. Targets include translation accuracy above 95%, delay below one second per sentence pause, voice DMOS above 4.2, tonal output and consistent performance across topics. It also requires secure feedback to BHASHINI for model refinement. These are requested benchmarks, not achieved results.