fix(asr): reject isolated_tail_closing hallucinations after phrase

fix(space): inline client_log route in main.py for single-file HF deploy

feat(api): register POST /client_log for HoloLens telemetry

fix(it): soften zlib compression gate; ultra-short tail after phrase

fix(forced-lang): reject ultra-short trimmed EN/IT segments by default

feat(it): warm-load transformers by default; soften RMS for fragments

fix: English segment word cap uses paired signals; fix Italian restore

fix(en): soften word-density reject with paired signals

tune(api-space): increase forced chunk context to 4.0s with tail trim enabled

feat(api-space): detect and trim weak late tail continuation sentences

tune(api-space): rebalance English confidence filters for realtime speech

fix(api-space): stricter English confidence+density rejection, less punitive Italian fallback

fix(api-space): disable aggressive forced segmentation by default for realtime demo

fix(api-space): reduce pre-decode over-filtering for short strong segments

fix(api-space): relax trimmed speech segment gate for EN/IT

feat(api-space): speech-segment-first forced decode with confidence-driven EN/IT acceptance

fix(api-space): suppress weak short forced-language tails after strong chunk

Revert "feat(api-space): generic forced-language tail acceptance filter (EN+IT)"

feat(api-space): generic forced-language tail acceptance filter (EN+IT)

Revert "fix(api-space): strip dangling THANK YOU/THANKS after .?!; trim silence for EN"

Revert "fix(api-space): apply dangling THANK YOU strip for auto/missing X-Forced-Language"

fix(api-space): apply dangling THANK YOU strip for auto/missing X-Forced-Language

fix(api-space): strip dangling THANK YOU/THANKS after .?!; trim silence for EN

feat(api-space): trim trailing silence on forced Italian chunks (HoloLens-style tails)

fix(api-space): force Whisper decode scores via GenerationConfig; strip dangling grazie tail after .?!

feat(api-space): expose Whisper transformers decode confidence metadata.

Enforce conservative forced-Italian acceptance and exact weak-tail reject logging.

Apply conservative confidence-based forced-Italian acceptance and weak-tail rejection.

Strip only dangling final English 'you' tail artifacts.

Strip only dangling final Italian conjunction tokens (E/ED).

Prioritize short Italian speech-RMS over VAD; add return/client response logs.

Use speech-window RMS (not full chunk RMS) for forced-Italian gating.

Add forced-Italian short-utterance mode with balanced RMS/min-chunk gates.

Tighten forced-Italian weak-tail rejection and apply real VAD pre-check.

Skip forced-Italian transformers warm-load to prevent HF build crash

Default forced-Italian to whisper-large-v3-turbo; short-chunk gate + VAD.

Remove Italian substring hallucination stripping; keep English-caption targets only.

Reduce EN/IT tail hallucinations: stricter thresholds, FW logprob, tail cleanup.

Strip redundant Italian Whisper tail "W e W" in forced-Italian sanitization.

Recover short forced-language utterances without hardcoded word rules.

Default forced-English CUDA path to quality model instead of realtime.

Use cross-model English consensus to suppress unstable trailing token hallucinations.

Gate forced-English realtime decodes by segment confidence before fallback selection.

Preserve short repeated utterances and avoid blanking tiny overlap chunks.

Retry empty forced-English chunks with a relaxed no-speech threshold.

Improve English ASR stability by rerunning suspicious fast-path decodes on the stronger EN model.

Revert "Default Italian production path to shared faster-whisper."

Default Italian production path to shared faster-whisper.

Improve Italian WER with accuracy-first forced decode defaults.

Reduce forced-Italian false blanks and hallucinated additions.