emoTag

emoTag helps you train emotion detection model for Chinese audios.

Environment

pip install -r requirement.txt

Data

We used Emotional Speech Dataset (ESD) for Speech Synthesis and Voice Conversion from HLT Singapore.

Train Emotion Classifier

Use this command to train a classifier. Adjust training setups in conf/logfbank_train-emo.json.

python train.py --config conf/logfbank_train-emo.json --name task_trial_1

Models and logs will be find in exp/.

usage: train.py [-h] [-c CONFIG] [-r RESUME] [-n NAME] [--lr LR] [--bs BS]
                [--train_utt2wav TRAIN_UTT2WAV] [--val_utt2wav VAL_UTT2WAV]
                [--blocks BLOCKS] [--optimizer OPTIMIZER]
                [--train_pad0 TRAIN_PAD0] [--devel_pad0 DEVEL_PAD0]
                [--pretrain PRETRAIN]

PyTorch Template

optional arguments:
  -h, --help            show this help message and exit
  -c CONFIG, --config CONFIG
                        config file path (default: None)
  -r RESUME, --resume RESUME
                        path to latest checkpoint (default: None)
  -n NAME, --name NAME
  --lr LR, --learning_rate LR
  --bs BS, --batch_size BS
  --train_utt2wav TRAIN_UTT2WAV
  --val_utt2wav VAL_UTT2WAV
  --blocks BLOCKS
  --optimizer OPTIMIZER
  --train_pad0 TRAIN_PAD0
  --devel_pad0 DEVEL_PAD0
  --pretrain PRETRAIN

Infer labels

python infer_label.py

Adjust the vad_file param and code if necessary to adapt to new tasks. infer_label.py adopted multiprocessing, increased cpu utilities rate and inference efficiency. See usage details below.

usage: infer_label.py [-h] [--vad_file VAD_FILE] [--model_dir MODEL_DIR]
                      [--data_dir DATA_DIR] [--output_dir OUTPUT_DIR]
                      [--process PROCESS]

parse model info

optional arguments:
  -h, --help            show this help message and exit
  --vad_file VAD_FILE
  --model_dir MODEL_DIR
  --data_dir DATA_DIR   data directory to be labelled
  --output_dir OUTPUT_DIR
  --process PROCESS     multiprocess number

infer_label.py automatically generates kill_label.sh script while running. Kill multiprocess labeling program simply by

bash kill_label.sh

Find inference failure log in bad_file.