Dionyssos's picture
fx w_axis
fd9e569
import typing
import types # fusion of forward() of Wav2Vec2
import gradio as gr
import matplotlib.pyplot as plt
import numpy as np
import spaces
import torch
import torch.nn as nn
from transformers import Wav2Vec2Processor
from transformers.models.wav2vec2.modeling_wav2vec2 import Wav2Vec2Model
from transformers.models.wav2vec2.modeling_wav2vec2 import Wav2Vec2PreTrainedModel
import audiofile
import audresample
device = 0 if torch.cuda.is_available() else "cpu"
duration = 2 # limit processing of audio
age_gender_model_name = "audeering/wav2vec2-large-robust-24-ft-age-gender"
expression_model_name = "audeering/wav2vec2-large-robust-12-ft-emotion-msp-dim"
class AgeGenderHead(nn.Module):
r"""Age-gender model head."""
def __init__(self, config, num_labels):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.dropout = nn.Dropout(config.final_dropout)
self.out_proj = nn.Linear(config.hidden_size, num_labels)
def forward(self, features, **kwargs):
x = features
x = self.dropout(x)
x = self.dense(x)
x = torch.tanh(x)
x = self.dropout(x)
x = self.out_proj(x)
return x
class AgeGenderModel(Wav2Vec2PreTrainedModel):
r"""Age-gender recognition model."""
def __init__(self, config):
super().__init__(config)
self.config = config
self.wav2vec2 = Wav2Vec2Model(config)
self.age = AgeGenderHead(config, 1)
self.gender = AgeGenderHead(config, 3)
self.init_weights()
def forward(
self,
frozen_cnn7,
):
hidden_states = self.wav2vec2(frozen_cnn7=frozen_cnn7) # runs only Transformer layers
hidden_states = torch.mean(hidden_states, dim=1)
logits_age = self.age(hidden_states)
logits_gender = torch.softmax(self.gender(hidden_states), dim=1)
return hidden_states, logits_age, logits_gender
# AgeGenderModel.forward() is switched to accept computed frozen CNN7 features from ExpressioNmodel
def _forward(
self,
frozen_cnn7=None, # CNN7 fetures of wav2vec2 calc. from CNN7 feature extractor (once)
attention_mask=None):
if attention_mask is not None:
# compute reduced attention_mask corresponding to feature vectors
attention_mask = self._get_feature_vector_attention_mask(
frozen_cnn7.shape[1], attention_mask, add_adapter=False
)
hidden_states, _ = self.wav2vec2.feature_projection(frozen_cnn7)
hidden_states = self.wav2vec2.encoder(
hidden_states,
attention_mask=attention_mask,
output_attentions=None,
output_hidden_states=None,
return_dict=None,
)[0]
return hidden_states
def _forward_and_cnn7(
self,
input_values,
attention_mask=None):
frozen_cnn7 = self.wav2vec2.feature_extractor(input_values)
frozen_cnn7 = frozen_cnn7.transpose(1, 2)
if attention_mask is not None:
# compute reduced attention_mask corresponding to feature vectors
attention_mask = self.wav2vec2._get_feature_vector_attention_mask(
frozen_cnn7.shape[1], attention_mask, add_adapter=False
)
hidden_states, _ = self.wav2vec2.feature_projection(frozen_cnn7) # grad=True non frozen
hidden_states = self.wav2vec2.encoder(
hidden_states,
attention_mask=attention_mask,
output_attentions=None,
output_hidden_states=None,
return_dict=None,
)[0]
return hidden_states, frozen_cnn7 #feature_proj is trainable thus we have to access the frozen_cnn7 before projection layer
class ExpressionHead(nn.Module):
r"""Expression model head."""
def __init__(self, config):
super().__init__()
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
self.dropout = nn.Dropout(config.final_dropout)
self.out_proj = nn.Linear(config.hidden_size, config.num_labels)
def forward(self, features, **kwargs):
x = features
x = self.dropout(x)
x = self.dense(x)
x = torch.tanh(x)
x = self.dropout(x)
x = self.out_proj(x)
return x
class ExpressionModel(Wav2Vec2PreTrainedModel):
r"""speech expression model."""
def __init__(self, config):
super().__init__(config)
self.config = config
self.wav2vec2 = Wav2Vec2Model(config)
self.classifier = ExpressionHead(config)
self.init_weights()
def forward(self, input_values):
hidden_states, frozen_cnn7 = self.wav2vec2(input_values)
hidden_states = torch.mean(hidden_states, dim=1)
logits = self.classifier(hidden_states)
return hidden_states, logits, frozen_cnn7
# Load models from hub
age_gender_processor = Wav2Vec2Processor.from_pretrained(age_gender_model_name)
age_gender_model = AgeGenderModel.from_pretrained(age_gender_model_name)
expression_processor = Wav2Vec2Processor.from_pretrained(expression_model_name)
expression_model = ExpressionModel.from_pretrained(expression_model_name)
# Emotion Calc. CNN features
age_gender_model.wav2vec2.forward = types.MethodType(_forward, age_gender_model)
expression_model.wav2vec2.forward = types.MethodType(_forward_and_cnn7, expression_model)
def process_func(x: np.ndarray, sampling_rate: int) -> typing.Tuple[str, dict, str]:
# batch audio
y = expression_processor(x, sampling_rate=sampling_rate)
y = y['input_values'][0]
y = y.reshape(1, -1)
y = torch.from_numpy(y).to(device)
# run through expression model
with torch.no_grad():
_, logits_expression, frozen_cnn7 = expression_model(y)
_, logits_age, logits_gender = age_gender_model(frozen_cnn7=frozen_cnn7)
# Plot A/D/V values
plot_expression(logits_expression[0, 0].item(), # implicit detach().cpu().numpy()
logits_expression[0, 1].item(),
logits_expression[0, 2].item())
expression_file = "expression.png"
plt.savefig(expression_file)
return (
f"{round(100 * logits_age[0, 0].item())} years", # age
{
"female": logits_gender[0, 0].item(),
"male": logits_gender[0, 1].item(),
"child": logits_gender[0, 2].item(),
},
expression_file,
)
@spaces.GPU
def recognize(input_file: str) -> typing.Tuple[str, dict, str]:
# sampling_rate, signal = input_microphone
# signal = signal.astype(np.float32, order="C") / 32768.0
if input_file is None:
raise gr.Error(
"No audio file submitted! "
"Please upload or record an audio file "
"before submitting your request."
)
signal, sampling_rate = audiofile.read(input_file, duration=duration)
# Resample to sampling rate supported byu the models
target_rate = 16000
signal = audresample.resample(signal, sampling_rate, target_rate)
return process_func(signal, target_rate)
def plot_expression_RIGID(arousal, dominance, valence):
r"""3D pixel plot of arousal, dominance, valence."""
# Voxels per dimension
voxels = 7
# Create voxel grid
x, y, z = np.indices((voxels + 1, voxels + 1, voxels + 1))
voxel = (
(x == round(arousal * voxels))
& (y == round(dominance * voxels))
& (z == round(valence * voxels))
)
projection = (
(x == round(arousal * voxels))
& (y == round(dominance * voxels))
& (z < round(valence * voxels))
)
colors = np.empty((voxel | projection).shape, dtype=object)
colors[voxel] = "#fcb06c"
colors[projection] = "#fed7a9"
ax = plt.figure().add_subplot(projection='3d')
ax.voxels(voxel | projection, facecolors=colors, edgecolor='k')
ax.set_xlim([0, voxels])
ax.set_ylim([0, voxels])
ax.set_zlim([0, voxels])
ax.set_aspect("equal")
ax.set_xlabel("arousal", fontsize="large", labelpad=0)
ax.set_ylabel("dominance", fontsize="large", labelpad=0)
ax.set_zlabel("valence", fontsize="large", labelpad=0)
ax.set_xticks(
list(range(voxels + 1)),
labels=[0, None, None, None, None, None, None, 1],
verticalalignment="bottom",
)
ax.set_yticks(
list(range(voxels + 1)),
labels=[0, None, None, None, None, None, None, 1],
verticalalignment="bottom",
)
ax.set_zticks(
list(range(voxels + 1)),
labels=[0, None, None, None, None, None, None, 1],
verticalalignment="top",
)
def explode(data):
"""
Expands a 3D array by creating gaps between voxels.
This function is used to create the visual separation between the voxels.
"""
shape_orig = np.array(data.shape)
shape_new = shape_orig * 2 - 1
retval = np.zeros(shape_new, dtype=data.dtype)
retval[::2, ::2, ::2] = data
return retval
def plot_expression(arousal, dominance, valence):
'''_h = cuda tensor (N_PIX, N_PIX, N_PIX)'''
N_PIX = 5
_h = np.random.rand(N_PIX, N_PIX, N_PIX) * 1e-3
adv = np.array([arousal, .994 - dominance, valence]).clip(0, .99)
arousal, dominance, valence = (adv * N_PIX).astype(np.int64) # find voxel
_h[arousal, dominance, valence] = .22
filled = np.ones((N_PIX, N_PIX, N_PIX), dtype=bool)
# upscale the above voxel image, leaving gaps
filled_2 = explode(filled)
# Shrink the gaps
x, y, z = np.indices(np.array(filled_2.shape) + 1).astype(float) // 2
x[1::2, :, :] += 1
y[:, 1::2, :] += 1
z[:, :, 1::2] += 1
ax = plt.figure().add_subplot(projection='3d')
f_2 = np.ones([2 * N_PIX - 1,
2 * N_PIX - 1,
2 * N_PIX - 1, 4], dtype=np.float64)
f_2[:, :, :, 3] = explode(_h)
cm = plt.get_cmap('cool')
f_2[:, :, :, :3] = cm(f_2[:, :, :, 3])[..., :3]
f_2[:, :, :, 3] = f_2[:, :, :, 3].clip(.01, .74)
print(f_2.shape, 'f_2 AAAA')
ecolors_2 = f_2
ax.voxels(x, y, z, filled_2, facecolors=f_2, edgecolors=.006 * ecolors_2)
ax.set_aspect('equal')
ax.set_zticks([0, N_PIX])
ax.set_xticks([0, N_PIX])
ax.set_yticks([0, N_PIX])
ax.set_zticklabels([f'{n/N_PIX:.2f}'[0:] for n in ax.get_zticks()])
ax.set_zlabel('valence', fontsize=10, labelpad=0)
ax.set_xticklabels([f'{n/N_PIX:.2f}' for n in ax.get_xticks()])
ax.set_xlabel('arousal', fontsize=10, labelpad=7)
# The y-axis rotation is corrected here from 275 to 90 degrees
ax.set_yticklabels([f'{1-n/N_PIX:.2f}' for n in ax.get_yticks()], rotation=90)
ax.set_ylabel('dominance', fontsize=10, labelpad=10)
ax.grid(False)
ax.plot([N_PIX, N_PIX], [0, N_PIX + .2], [N_PIX, N_PIX], 'g', linewidth=1)
ax.plot([0, N_PIX], [N_PIX, N_PIX + .24], [N_PIX, N_PIX], 'k', linewidth=1)
# Missing lines on the top face
ax.plot([0, 0], [0, N_PIX], [N_PIX, N_PIX], 'darkred', linewidth=1)
ax.plot([0, N_PIX], [0, 0], [N_PIX, N_PIX], 'darkblue', linewidth=1)
# Set pane colors after plotting the lines
# UPDATED: Replaced `w_xaxis` with `xaxis` and `w_yaxis` with `yaxis`.
ax.xaxis.set_pane_color((0.8, 0.8, 0.8, 0.5))
ax.yaxis.set_pane_color((0.8, 0.8, 0.8, 0.5))
ax.zaxis.set_pane_color((0.8, 0.8, 0.8, 0.0))
# Restore the limits to prevent the plot from expanding
ax.set_xlim(0, N_PIX)
ax.set_ylim(0, N_PIX)
ax.set_zlim(0, N_PIX)
#plt.show()
# ------
description = (
"Estimate **age**, **gender**, and **expression** "
"of the speaker contained in an audio file or microphone recording. \n"
f"The model [{age_gender_model_name}]"
f"(https://huggingface.co/{age_gender_model_name}) "
"recognises age and gender, "
f"whereas [{expression_model_name}]"
f"(https://huggingface.co/{expression_model_name}) "
"recognises the expression dimensions arousal, dominance, and valence. "
)
with gr.Blocks() as demo:
with gr.Tab(label="Speech analysis"):
with gr.Row():
with gr.Column():
gr.Markdown(description)
input = gr.Audio(
sources=["upload", "microphone"],
type="filepath",
label="Audio input",
min_length=0.025, # seconds
)
gr.Examples(
[
"female-46-neutral.wav",
"female-20-happy.wav",
"male-60-angry.wav",
"male-27-sad.wav",
],
[input],
label="Examples from CREMA-D, ODbL v1.0 license",
)
gr.Markdown("Only the first two seconds of the audio will be processed.")
submit_btn = gr.Button(value="Submit")
with gr.Column():
output_age = gr.Textbox(label="Age")
output_gender = gr.Label(label="Gender")
output_expression = gr.Image(label="Expression")
outputs = [output_age, output_gender, output_expression]
submit_btn.click(recognize, input, outputs)
demo.launch(debug=True)