Context
The first fluidstreamd implementation used FluidAudio StreamingEouAsrManager (Parakeet EOU 120M, English-only, 160ms chunks, real end-of-utterance finals, sub-second). It was dropped in favor of Nemotron multilingual (ru+en, 2s tier) to support Russian.
Goal
Re-add the EOU engine behind --engine eou for English-only meetings that want lower latency (160ms) and automatic mid-stream EOU finals, while Nemotron multilingual stays the default.
Notes
- EOU model is lowercase / no punctuation.
- Trivial: the earlier code path exists in git history (pre-Nemotron commit); select the manager by flag.
Context
The first fluidstreamd implementation used FluidAudio
StreamingEouAsrManager(Parakeet EOU 120M, English-only, 160ms chunks, real end-of-utterance finals, sub-second). It was dropped in favor of Nemotron multilingual (ru+en, 2s tier) to support Russian.Goal
Re-add the EOU engine behind
--engine eoufor English-only meetings that want lower latency (160ms) and automatic mid-stream EOU finals, while Nemotron multilingual stays the default.Notes