| # ############################################################################ | |
| # Model: WAV2VEC base for Emotion Recognition | |
| # ############################################################################ | |
| # Hparams NEEDED | |
| HPARAMS_NEEDED: ["encoder_dim", "out_n_neurons", "label_encoder", "softmax"] | |
| # Modules Needed | |
| MODULES_NEEDED: ["wav2vec2", "avg_pool", "output_mlp"] | |
| # Feature parameters | |
| wav2vec2_hub: microsoft/wavlm-base-plus-sv | |
| # Pretrain folder (HuggingFace) | |
| pretrained_path: Porjaz/wavlm-base-emo-fi | |
| # parameters | |
| encoder_dim: 768 | |
| out_n_neurons: 5 | |
| wav2vec2: !new:speechbrain.lobes.models.huggingface_transformers.wav2vec2.Wav2Vec2 | |
| source: !ref <wav2vec2_hub> | |
| output_norm: True | |
| freeze: True | |
| save_path: wav2vec2_checkpoints | |
| avg_pool: !new:speechbrain.nnet.pooling.StatisticsPooling | |
| return_std: False | |
| label_lin: !new:speechbrain.nnet.linear.Linear | |
| input_size: !ref <encoder_dims> | |
| n_neurons: !ref <n_classes> | |
| bias: False | |
| model: !new:torch.nn.ModuleList | |
| - [!ref <label_lin>] | |
| modules: | |
| wav2vec2: !ref <wav2vec2> | |
| label_lin: !ref <label_lin> | |
| avg_pool: !ref <avg_pool> | |
| softmax: !new:speechbrain.nnet.activations.Softmax | |
| label_encoder: !new:speechbrain.dataio.encoder.CategoricalEncoder | |
| pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer | |
| loadables: | |
| wav2vec2: !ref <wav2vec2> | |
| model: !ref <model> | |
| label_encoder: !ref <label_encoder> | |
| paths: | |
| wav2vec2: !ref <pretrained_path>/wav2vec2.ckpt | |
| model: !ref <pretrained_path>/model.ckpt | |
| label_encoder: !ref <pretrained_path>/label_encoder.txt | |