configs/infer.yaml

# Inference Configuration

# Model
model:
  checkpoint: "model/vulehuubinh"
  name: "nguyenvulebinh/wav2vec2-base-vi-vlsp2020"
  head_hidden_dim: 256

# Audio Processing
audio:
  sampling_rate: 16000
  max_duration: 5

# Inference
inference:
  batch_size: 1
  device: "cuda"

# Input
input:
  audio_path: null
  audio_dir: null

# Output
output:
  dir: "output/predictions"
  save_results: true
  format: "json"

# Label Mappings
# NOTE: Model was trained with Female=0, Male=1 (opposite of finetune.yaml order)
# This is because pandas .map() may have processed labels in different order
labels:
  gender:
    0: "Female"
    1: "Male"
  dialect:
    0: "North"
    1: "Central"
    2: "South"