remove compute_performance.py

.gitignore

@@ -1 +1,2 @@
-*.feather
+*.feather
+*.py

compute_performance.py

@@ -1,99 +0,0 @@
-import numpy as np
-import pandas as pd
-from sklearn.metrics import roc_auc_score
-
-import pandas as pd
-from concurrent.futures import ProcessPoolExecutor
-from tqdm import tqdm
-from silero_vad import  read_audio, get_speech_timestamps
-from functools import partial
-
-from pathlib import Path
-str(Path().resolve() / "silero-vad/src/silero-vad")
-from silero_vad import utils_vad
-# from utils_vad import init_jit_model, OnnxWrapper
-import torch
-torch.set_num_threads(1)
-
-def load_silero_vad(onnx=False, model_file_path=None):
-    if onnx:
-        model = utils_vad.OnnxWrapper(model_file_path, force_onnx_cpu=True)
-    else:
-        model = utils_vad.init_jit_model(model_file_path)
-    
-    return model
-
-def init_worker(model_file_path):
-    """Initialize the model inside each worker process"""
-    global model
-    model = load_silero_vad(onnx=False, model_file_path=model_file_path)
-
-
-
-def get_vad(file, threshold):
-    if pd.isna(file):
-        return None
-    wav = read_audio(file)
-    speech_timestamps = get_speech_timestamps(
-        wav,
-        model,
-        return_seconds=True,
-        threshold=threshold
-    )
-    return speech_timestamps
-
-def process_vad_parallel(df, threshold, column_name, model_file_path):
-    results = []
-    with ProcessPoolExecutor(max_workers=8, initializer=partial(init_worker, model_file_path) ) as executor:
-        futures = {executor.submit(get_vad, file, threshold): i for i, file in enumerate(df["audio_path"])}
-        for future in tqdm(futures, total=len(df), desc=f"Processing {column_name}"):
-            results.append(future.result())  # Collect results
-    df[column_name] = results
-    return df
-
-
-
-def create_frame_labels(segments, duration, frame_size=0.01):
-    frames = np.zeros(int(duration / frame_size))
-    for seg in segments:
-        start_idx = int(seg['start'] / frame_size)
-        end_idx = int(seg['end'] / frame_size)
-        frames[start_idx:end_idx] = 1
-    return frames
-
-def compute_auc_roc(df, actual_col, predicted_col, frame_size=0.01):
-    max_time = max(
-        max(seg['end'] for row in df[actual_col] for seg in row),
-        max(seg['end'] for row in df[predicted_col] for seg in row)
-    )
-    
-    gt_labels = create_frame_labels([seg for row in df[actual_col] for seg in row], max_time, frame_size)
-    pred_labels = create_frame_labels([seg for row in df[predicted_col] for seg in row], max_time, frame_size)
-    
-    auc_roc = roc_auc_score(gt_labels, pred_labels)
-    return auc_roc
-
-
-
-df = pd.read_feather("./val.feather")
-model_file_path = "/home/sourabh/Desktop/dev/hum-vad/HumAware-VAD/humaware_vad.jit"
-df = process_vad_parallel(df, 0.5, "unhum_vad_output_0.5", model_file_path=model_file_path)
-df = process_vad_parallel(df, 0.9, "unhum_vad_output_0.9", model_file_path=model_file_path)
-
-model_file_path = "/home/sourabh/Desktop/dev/hum-vad/.venv/lib/python3.12/site-packages/silero_vad/data/silero_vad.jit"
-df = process_vad_parallel(df, 0.5, "silero_vad_output_0.5", model_file_path=model_file_path)
-df = process_vad_parallel(df, 0.9, "silero_vad_output_0.9", model_file_path=model_file_path)
-
-
-
-auc_roc_score = compute_auc_roc(df, "speech_ts", "unhum_vad_output_0.5")
-print(f"AUC-ROC Score: {auc_roc_score:.4f}")
-
-auc_roc_score = compute_auc_roc(df, "speech_ts", "unhum_vad_output_0.9")
-print(f"AUC-ROC Score unhum_vad_output_0.9: {auc_roc_score:.4f}")
-
-auc_roc_score = compute_auc_roc(df, "speech_ts", "silero_vad_output_0.5")
-print(f"AUC-ROC Score silero_vad_output_0.5: {auc_roc_score:.4f}")
-
-auc_roc_score = compute_auc_roc(df, "speech_ts", "silero_vad_output_0.9")
-print(f"AUC-ROC Score silero_vad_output_0.9: {auc_roc_score:.4f}")