ckpt download speed

README.md

@@ -12,7 +12,7 @@ tags:
 
 # Wav2Small2.0 - Arousal / Dominance / Valence
 
-Please note that this model is for research purpose only. A commercial [license](https://www.audeering.com/products/devaice/) can be acquired with audEERING. The model expects a raw audio signal 16KHz as input, and outputs: arousal, dominance valence in range [0, 1], as well as Anger/Happiness/Neutral/Sad probability. The model is created following the [Wav2Small paper](https://arxiv.org/abs/2408.13920) and has a total of 17K params.
+Please note that this model is for research purpose only. A commercial [license](https://www.audeering.com/products/devaice/) can be acquired with audEERING. The model expects a raw audio signal 16KHz as input, and outputs: arousal, dominance valence in range [0, 1]. The model is created following the [Wav2Small paper](https://arxiv.org/abs/2408.13920) and has a total of 17K params.
 
 
 # How To
@@ -20,11 +20,10 @@ Please note that this model is for research purpose only. A commercial [license]
 ```python
 import torch
 import numpy as np
-import torch.nn.functional as F
 import librosa
-from transformers.models.wav2vec2.modeling_wav2vec2 import Wav2Vec2PreTrainedModel
+from transformers import Wav2Vec2PreTrainedModel, PretrainedConfig
 from torch import nn
-from transformers import PretrainedConfig
+
 
 
 
@@ -97,7 +96,7 @@ class Spectrogram(nn.Module):
 
         real = self.conv_real(x)
         imag = self.conv_imag(x)
-        return real ** 2 + imag ** 2  # bs, mel, time-frames
+        return real ** 2 + imag ** 2  # bs, freq, time-frames
 
 
 class LogmelFilterBank(nn.Module):