# models_config.yaml
version: 1

app:
  max_duration_sec: 900        # hard cap enforced by UI
  sample_rate_hz: 16000
  save_intermediate_audio: true   # enhanced wav
  save_tables: true
  retention_runs: 20              # keep last N runs, clean older on startup

paths:
  cache_dir: .cache
  output_dir: outputs

preprocess:
  enhance_enabled: true
  enhance_backend: demucs        # demucs | rnnoise
  enhance_strength: 0.55         # 0.0 to 1.0
  chunk_sec: 4.0
  chunk_overlap_sec: 0.10
  loudness_target_lufs: -16

vad:
  enabled: true
  backend: silero                # future proof
  threshold: 0.5                 # 0..1 higher = stricter
  min_speech_ms: 250
  min_silence_ms: 150
  speech_pad_ms: 80
  merge_gap_ms: 200

asr:
  model_id: Systran/faster-whisper-large-v3
  language: auto
  word_timestamps: true
  compute_type: auto
  vad_filter: false



diarization:
  enable: true
  backend: pyannote
  min_speakers: 2
  max_speakers: 8
  overlap_handling: assign_max_overlap
  allow_unknown: true

reporting:
  round_seconds: 3               # decimals for start/end
  round_confidence: 3
  include_device_info: true