File size: 6,719 Bytes
859f0be |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 |
from transformers import Pipeline
import numpy as np
import torch
from typing import Dict, Union, List, Optional
from pathlib import Path
import logging
from datasets import Dataset
logger = logging.getLogger(__name__)
class ProsodyEmbeddingPipeline(Pipeline):
def __init__(
self,
speaker_stats,
f0_interp,
f0_normalize,
stats_dir: Optional[str] = None,
**kwargs
):
super().__init__(**kwargs)
self.stats_dir = Path(stats_dir) if stats_dir else None
self.speaker_stats = speaker_stats
self.f0_interp = f0_interp
self.f0_normalize = f0_normalize
def _sanitize_parameters(self, **kwargs):
preprocess_kwargs = {}
forward_kwargs = {}
postprocess_kwargs = {
"return_tensors": kwargs.pop("return_tensors", "pt")
}
return preprocess_kwargs, forward_kwargs, postprocess_kwargs
def preprocess(self, inputs: Union[str, Dict, Dataset]) -> Dict:
"""Preprocess inputs"""
spkr_id = inputs['speaker_id']
stats = self.speaker_stats[spkr_id]
if self.f0_interp:
f0 = torch.Tensor(inputs['f0_interp'])
else:
f0 = torch.Tensor(inputs['f0'])
f0_orig = f0.clone() # Save original f0 before normalization
intensity = torch.Tensor(inputs['intensity'])
intensity_orig = intensity.clone() # Save original intensity before normalization
if self.f0_normalize:
ii = f0 != 0
if stats.f0_std != 0:
f0[ii] = (f0[ii] - stats.f0_mean) / stats.f0_std
intensity_ii = intensity != 0
if stats.intensity_std != 0:
intensity[intensity_ii] = (intensity[intensity_ii] - stats.intensity_mean) / stats.intensity_std
if not self.f0_interp:
zero_indices = f0 == 0
zero_mask = zero_indices * 1.0
inputs = {
'f0': f0,
'intensity': intensity,
'zero_mask': zero_mask if self.f0_normalize and not self.f0_interp else None,
'f0_mean': stats.f0_mean,
'f0_std': stats.f0_std,
'intensity_mean': stats.intensity_mean,
'intensity_std': stats.intensity_std,
'f0_orig': f0_orig, # original features before normalization
'intensity_orig': intensity_orig, # original features
'speaker_id': spkr_id,
}
return inputs
def _forward(self, features: Dict) -> Dict:
"""Run the model on the preprocessed features"""
self.model.eval()
f0 = torch.Tensor(features['f0'])
intensity = torch.Tensor(features['intensity'])
if self.f0_interp:
stacked_features = torch.stack([f0, intensity], dim=0).to(self.model.device)
else:
zero_mask = torch.Tensor(features['zero_mask'])
stacked_features = torch.stack([f0, intensity, zero_mask], dim=0).to(self.model.device)
stacked_features = stacked_features.unsqueeze(0)
with torch.no_grad():
model_outputs = self.model(features=stacked_features)
outputs = {
**model_outputs,
'input_features': {
'zero_mask': zero_mask if self.f0_normalize and not self.f0_interp else None,
'f0_orig': features['f0_orig'],
'f0_mean': features['f0_mean'],
'f0_std': features['f0_std'],
'intensity_mean': features['intensity_mean'],
'intensity_std': features['intensity_std'],
'intensity_orig': features['intensity_orig']
}
}
return outputs
def postprocess(self, outputs: Dict, return_tensors: str = "pt") -> Dict:
"""Convert outputs to the desired format and calculate metrics"""
input_f0 = outputs['input_features']['f0_orig']
output_f0 = outputs['f0'][0,:,:]
f0_recon = output_f0
# revert normalization
if self.f0_normalize:
f0_recon[0] = f0_recon[0] * outputs['input_features']["f0_std"] + outputs['input_features']["f0_mean"]
if not self.f0_interp:
mask = torch.where(f0_recon[2] < 0.5, torch.tensor([1.0]), torch.tensor([0.0]))
f0_recon[0] = (f0_recon[0] * mask)
f0_recon[1] = f0_recon[1] * outputs['input_features']["intensity_std"] + outputs['input_features']["intensity_mean"]
epsilon = 1e-10
DIFF_THRESHOLD = 0.2
# F0 metrics calculation
input_f0_np = input_f0.cpu().numpy()
output_f0_np = f0_recon[0].cpu().numpy() # Use f0_recon[0] instead of output_f0
# Truncate both arrays to multiple of 16
length = len(input_f0_np)
truncated_length = (length // 16) * 16
input_f0_np = input_f0_np[:truncated_length]
output_f0_np = output_f0_np[:truncated_length]
input_f0_safe = np.where(input_f0_np == 0, epsilon, input_f0_np)
rel_diff = np.abs(input_f0_np - output_f0_np) / np.abs(input_f0_safe)
diff_points = rel_diff > DIFF_THRESHOLD
diff_count = np.sum(diff_points)
total_points = len(input_f0_np)
f0_large_diff_percent = (diff_count / total_points) * 100
# intensity metrics calculation
input_intensity_np = outputs['input_features']['intensity_orig'].cpu().numpy()
output_intensity_np = f0_recon[1].cpu().numpy()
# Truncate intensity arrays to multiple of 16
length = len(input_intensity_np)
truncated_length = (length // 16) * 16
input_intensity_np = input_intensity_np[:truncated_length]
output_intensity_np = output_intensity_np[:truncated_length]
intensity_rmse = np.sqrt(np.mean((input_intensity_np - output_intensity_np) ** 2))
outputs['f0_recon'] = output_f0_np
outputs['intensity_recon'] = output_intensity_np
# Add metrics to outputs
outputs['metrics'] = {
'f0_large_diff_percent': f0_large_diff_percent.item(),
'intensity_rmse': float(intensity_rmse)
}
print(f"outputs['metrics']", outputs['metrics'])
if return_tensors == "np":
outputs = {
k: v.cpu().numpy() if torch.is_tensor(v) else v
for k, v in outputs.items()
}
return outputs |