scripts/bash/misc/eval_audio.sh

#!/bin/bash

export TOKENIZERS_PARALLELISM="false"
export DECORD_DUPLICATE_WARNING_THRESHOLD=1.0

METRICS="audio-score" 
RESULTS_DIR="./evaluation_results/audio_eval"
INFER_DATA_ROOT="samples/audio_eval"

declare -a DATASETS=("audiocaps" "javismini")  #  
declare -a MAX_AUDIO_LENS=(8.0 4.0) # 

for model in "Wan2.1"; do  # "baseline" 

for folder in "${INFER_DATA_ROOT}"/"${model}"/*; do
method=$(basename "$folder")

for subfolder in "${folder}"/ep*; do
ep=$(basename "$subfolder")

for i in "${!DATASETS[@]}"; do
dataset=${DATASETS[$i]}
max_audio_len_s=${MAX_AUDIO_LENS[$i]}
gpu=0

# model="baseline"
# method="audioldm2"
# ep="ep-1"
# dataset="javismini"
# max_audio_len_s="4.0"

INPUT_FILE="data/meta/audio_eval/${dataset}.csv"
INFER_DATA_DIR="${INFER_DATA_ROOT}/${model}/${method}/${ep}/${dataset}"
OUTPUT_FILE_PATH="${RESULTS_DIR}/${model}_${method}_${ep}_${dataset}.json"

export CUDA_VISIBLE_DEVICES=${gpu} 
torchrun --nproc_per_node=1 -m eval.javisbench.main \
    --input_file "${INPUT_FILE}" \
    --infer_data_dir "${INFER_DATA_DIR}" \
    --output_file "${OUTPUT_FILE_PATH}" \
    --max_audio_len_s "${max_audio_len_s}" \
    --metrics ${METRICS}

done

done

done

done