fix: reduce num_pipelines 3→1 — Parakeet OOM on 8GB CPU

trigger: rebuild Space to fix 503 (pipelines stuck)

revert: Qwen3-TTS → Kokoro (Qwen3 1.7B fails on CPU)

feat: switch TTS Kokoro → Qwen3-TTS (matching production stack)

feat: switch to production stack — Parakeet STT + aggressive VAD

tune: VAD silence 2000ms → 2500ms (more breathing room)

tune: lower VAD silence from 3000ms to 2000ms (2s feels more responsive)

fix: correct hyphen→underscore in --no_enable_live_transcription flag

fix: disable live transcription to prevent progressive STT truncation

Fix VAD cutoff: increase min_silence_ms to 3000ms, add continuation/reopen settings

fix: VAD patch v3 — use Python to locate file, sed with correct comma pattern

fix: revert to working Dockerfile (d7feaf9) — VAD patches breaking build

fix: use Python instead of find/sed to patch VAD (build was failing)

fix: patch vad_handler.py default from 64ms to 800ms (was patching wrong file)

fix: patch realtime VAD default from 300ms to 800ms silence

fix: increase VAD silence to 800ms, slow TTS to 0.85x, lower threshold to 0.4

fix: increase VAD silence timeout from 64ms to 500ms

Revert "feat: add local uncensored LLM endpoint (/v1/chat/completions) via llama-cpp-python"

feat: add local uncensored LLM endpoint (/v1/chat/completions) via llama-cpp-python

feat: add --kokoro_speed 1.0 parameter

fix: use shell form CMD so HF_TOKEN env var is expanded

revert: back to responses-api (token now valid)

fix: switch to chat-completions backend for Cerebras compatibility

fix: use --num_pipelines 3 to allow 3 concurrent sessions

fix: revert invalid --pool_size flag that crashed Space

fix: increase pool_size to 3 to prevent session slot exhaustion

fix: add kokoro>=0.9.2 and soundfile for Kokoro TTS handler

Fix: pip install faster-whisper for CPU STT

Fix: use faster-whisper (CPU) + --stt_device cpu for HF Spaces

Fix: use --ws_host/--ws_port instead of --host/--port

Production Dockerfile: Kokoro TTS + faster-whisper STT + Cerebras LLM

initial commit

Duplicate from posit/shiny-for-python-template