remove compute_performance.py
Browse files- .gitignore +2 -1
- compute_performance.py +0 -99
.gitignore
CHANGED
@@ -1 +1,2 @@
|
|
1 |
-
*.feather
|
|
|
|
1 |
+
*.feather
|
2 |
+
*.py
|
compute_performance.py
DELETED
@@ -1,99 +0,0 @@
|
|
1 |
-
import numpy as np
|
2 |
-
import pandas as pd
|
3 |
-
from sklearn.metrics import roc_auc_score
|
4 |
-
|
5 |
-
import pandas as pd
|
6 |
-
from concurrent.futures import ProcessPoolExecutor
|
7 |
-
from tqdm import tqdm
|
8 |
-
from silero_vad import read_audio, get_speech_timestamps
|
9 |
-
from functools import partial
|
10 |
-
|
11 |
-
from pathlib import Path
|
12 |
-
str(Path().resolve() / "silero-vad/src/silero-vad")
|
13 |
-
from silero_vad import utils_vad
|
14 |
-
# from utils_vad import init_jit_model, OnnxWrapper
|
15 |
-
import torch
|
16 |
-
torch.set_num_threads(1)
|
17 |
-
|
18 |
-
def load_silero_vad(onnx=False, model_file_path=None):
|
19 |
-
if onnx:
|
20 |
-
model = utils_vad.OnnxWrapper(model_file_path, force_onnx_cpu=True)
|
21 |
-
else:
|
22 |
-
model = utils_vad.init_jit_model(model_file_path)
|
23 |
-
|
24 |
-
return model
|
25 |
-
|
26 |
-
def init_worker(model_file_path):
|
27 |
-
"""Initialize the model inside each worker process"""
|
28 |
-
global model
|
29 |
-
model = load_silero_vad(onnx=False, model_file_path=model_file_path)
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
def get_vad(file, threshold):
|
34 |
-
if pd.isna(file):
|
35 |
-
return None
|
36 |
-
wav = read_audio(file)
|
37 |
-
speech_timestamps = get_speech_timestamps(
|
38 |
-
wav,
|
39 |
-
model,
|
40 |
-
return_seconds=True,
|
41 |
-
threshold=threshold
|
42 |
-
)
|
43 |
-
return speech_timestamps
|
44 |
-
|
45 |
-
def process_vad_parallel(df, threshold, column_name, model_file_path):
|
46 |
-
results = []
|
47 |
-
with ProcessPoolExecutor(max_workers=8, initializer=partial(init_worker, model_file_path) ) as executor:
|
48 |
-
futures = {executor.submit(get_vad, file, threshold): i for i, file in enumerate(df["audio_path"])}
|
49 |
-
for future in tqdm(futures, total=len(df), desc=f"Processing {column_name}"):
|
50 |
-
results.append(future.result()) # Collect results
|
51 |
-
df[column_name] = results
|
52 |
-
return df
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
def create_frame_labels(segments, duration, frame_size=0.01):
|
57 |
-
frames = np.zeros(int(duration / frame_size))
|
58 |
-
for seg in segments:
|
59 |
-
start_idx = int(seg['start'] / frame_size)
|
60 |
-
end_idx = int(seg['end'] / frame_size)
|
61 |
-
frames[start_idx:end_idx] = 1
|
62 |
-
return frames
|
63 |
-
|
64 |
-
def compute_auc_roc(df, actual_col, predicted_col, frame_size=0.01):
|
65 |
-
max_time = max(
|
66 |
-
max(seg['end'] for row in df[actual_col] for seg in row),
|
67 |
-
max(seg['end'] for row in df[predicted_col] for seg in row)
|
68 |
-
)
|
69 |
-
|
70 |
-
gt_labels = create_frame_labels([seg for row in df[actual_col] for seg in row], max_time, frame_size)
|
71 |
-
pred_labels = create_frame_labels([seg for row in df[predicted_col] for seg in row], max_time, frame_size)
|
72 |
-
|
73 |
-
auc_roc = roc_auc_score(gt_labels, pred_labels)
|
74 |
-
return auc_roc
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
df = pd.read_feather("./val.feather")
|
79 |
-
model_file_path = "/home/sourabh/Desktop/dev/hum-vad/HumAware-VAD/humaware_vad.jit"
|
80 |
-
df = process_vad_parallel(df, 0.5, "unhum_vad_output_0.5", model_file_path=model_file_path)
|
81 |
-
df = process_vad_parallel(df, 0.9, "unhum_vad_output_0.9", model_file_path=model_file_path)
|
82 |
-
|
83 |
-
model_file_path = "/home/sourabh/Desktop/dev/hum-vad/.venv/lib/python3.12/site-packages/silero_vad/data/silero_vad.jit"
|
84 |
-
df = process_vad_parallel(df, 0.5, "silero_vad_output_0.5", model_file_path=model_file_path)
|
85 |
-
df = process_vad_parallel(df, 0.9, "silero_vad_output_0.9", model_file_path=model_file_path)
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
auc_roc_score = compute_auc_roc(df, "speech_ts", "unhum_vad_output_0.5")
|
90 |
-
print(f"AUC-ROC Score: {auc_roc_score:.4f}")
|
91 |
-
|
92 |
-
auc_roc_score = compute_auc_roc(df, "speech_ts", "unhum_vad_output_0.9")
|
93 |
-
print(f"AUC-ROC Score unhum_vad_output_0.9: {auc_roc_score:.4f}")
|
94 |
-
|
95 |
-
auc_roc_score = compute_auc_roc(df, "speech_ts", "silero_vad_output_0.5")
|
96 |
-
print(f"AUC-ROC Score silero_vad_output_0.5: {auc_roc_score:.4f}")
|
97 |
-
|
98 |
-
auc_roc_score = compute_auc_roc(df, "speech_ts", "silero_vad_output_0.9")
|
99 |
-
print(f"AUC-ROC Score silero_vad_output_0.9: {auc_roc_score:.4f}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|