clean

README.md

@@ -44,15 +44,14 @@ Florian Eyben, Felix Burkhardt, Björn Schuller.
 # Usage
 ```python
 from transformers import AutoModelForAudioClassification
+from transformers.models.wav2vec2.modeling_wav2vec2 import (Wav2Vec2Model,
+    Wav2Vec2PreTrainedModel)
 import torch
 import types
 import torch.nn as nn
-from transformers.models.wav2vec2.modeling_wav2vec2 import (Wav2Vec2Model,
-    Wav2Vec2PreTrainedModel)
 
 # speech signal 16 KHz
-sampling_rate = 16000
-signal = torch.zeros((1, sampling_rate))
+signal = torch.rand((1, 16000))
 device = 'cpu'
 
 class RegressionHead(nn.Module):
@@ -67,8 +66,6 @@ class RegressionHead(nn.Module):
         return self.out_proj(x.tanh())
 
 class Dawn(Wav2Vec2PreTrainedModel):
-    r"""https://arxiv.org/abs/2203.07378"""
-
     def __init__(self, config):
 
         super().__init__(config)
@@ -83,7 +80,7 @@ class Dawn(Wav2Vec2PreTrainedModel):
         x = self.wav2vec2(x / variance.sqrt())[0]
         return self.classifier(x.mean(1)).clip(0, 1)
 
-def _infer(self, x):
+def _forward(self, x):
     '''x: (batch, audio-samples-16KHz)'''
     x = (x + self.config.mean) / self.config.std  # plus
     x = self.ssl_model(x, attention_mask=None).last_hidden_state
@@ -99,22 +96,25 @@ def _infer(self, x):
         ], 1)
     return self.ser_model(x)
 
+
 # WavLM
+
 base = AutoModelForAudioClassification.from_pretrained(
-    '3loi/SER-Odyssey-Baseline-WavLM-Multi-Attributes',
-    trust_remote_code=True  # fun definitions see 3loi/SER-.. repo
-).to(device).eval()
-base.forward = types.MethodType(_infer, base)
+  '3loi/SER-Odyssey-Baseline-WavLM-Multi-Attributes',
+  trust_remote_code=True).to(device).eval()
+base.forward = types.MethodType(_forward, base)
 
 # Wav2Vec2.0
+
 dawn = Dawn.from_pretrained(
-    'audeering/wav2vec2-large-robust-12-ft-emotion-msp-dim'
+  'audeering/wav2vec2-large-robust-12-ft-emotion-msp-dim'
 ).to(device).eval()
 
 
 def wav2small(x):
     return .5 * dawn(x) + .5 * base(x)
 
+
 with torch.no_grad():
     pred = wav2small(signal.to(device))
 print(f'\nArousal  = {pred[0, 0]} Dominance= {pred[0, 1]}',