update

data/early_media_examples_wav.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66ac7e2aa7f8fe0b001038f3041b9e8d03a27f30d99c27dab2efe82c93fc6c0f
+size 27919350

examples/channel_dual_to_single.py

@@ -15,7 +15,7 @@ def get_args():
 
     parser.add_argument(
         "--wav_dir",
-        default=(project_path / "data/early_media/60/wav").as_posix(),
+        default=(project_path / "data/early_media/63/wav").as_posix(),
         type=str
     )
 
@@ -43,6 +43,8 @@ def main():
             continue
 
         # print(signal.shape)
+        if signal.ndim != 2:
+            continue
         signal = signal[:, 0]
         # print(signal.shape)
 

examples/early_media_and_voicemail/step_1_early_media_and_voicemail.py

@@ -0,0 +1,182 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+import os
+from pathlib import Path
+import sys
+import tempfile
+import zipfile
+
+pwd = os.path.abspath(os.path.dirname(__file__))
+sys.path.append(os.path.join(pwd, "../../"))
+
+import numpy as np
+import pandas as pd
+from scipy.io import wavfile
+import torch
+from tqdm import tqdm
+
+from project_settings import project_path
+from toolbox.torch.utils.data.vocabulary import Vocabulary
+
+
+area_code = 886
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--voicemail_model_file",
+        # default="D:/Users/tianx/VoiceProxyProjects/VoicemailDetection/trained_models/cnn_voicemail_ms_my_20240122.zip",
+        # default="D:/Users/tianx/VoiceProxyProjects/VoicemailDetection/trained_models/cnn_voicemail_th_th_20221107.zip",
+        default="D:/Users/tianx/VoiceProxyProjects/VoicemailDetection/trained_models/cnn_voicemail_zh_tw_20230814.zip",
+        # default=(project_path / "trained_models").as_posix(),
+        type=str
+    )
+    parser.add_argument(
+        "--calling_wav_dir",
+        default=(project_path / "data/calling/{area_code}/wav".format(area_code=area_code)).as_posix(),
+        type=str
+    )
+    parser.add_argument(
+        "--early_media_wav_dir",
+        default=(project_path / "data/early_media/{area_code}/wav".format(area_code=area_code)).as_posix(),
+        type=str
+    )
+    parser.add_argument("--output_file", default="result.xlsx", type=str)
+
+    args = parser.parse_args()
+    return args
+
+
+class VoicemailDetection(object):
+    def __init__(self, model_file: str):
+        self.model_file = model_file
+
+        self.model = None
+        self.vocabulary = None
+        self.load_models()
+
+    def load_models(self):
+        model_file = Path(self.model_file)
+        model_name = model_file.stem
+
+        with zipfile.ZipFile(model_file.as_posix(), "r") as f_zip:
+            out_root = Path(tempfile.gettempdir()) / "cnn_voicemail"
+            out_root.mkdir(parents=True, exist_ok=True)
+            f_zip.extractall(path=out_root)
+
+        tgt_path = out_root / model_name
+        pth_path = tgt_path / "cnn_voicemail.pth"
+        vocab_path = tgt_path / "vocabulary"
+
+        with open(pth_path.as_posix(), "rb") as f:
+            model = torch.jit.load(f, map_location="cpu")
+        vocabulary = Vocabulary.from_files(vocab_path.as_posix())
+        self.model = model
+        self.vocabulary = vocabulary
+
+    def infer(self, inputs: np.ndarray):
+        # infer
+        inputs = inputs / (1 << 15)
+        inputs = torch.tensor(inputs, dtype=torch.float32)
+        inputs = torch.unsqueeze(inputs, dim=0)
+
+        outputs = self.model(inputs)
+
+        probs = outputs["probs"]
+        argmax = torch.argmax(probs, dim=-1)
+        probs = probs.tolist()[0]
+        argmax = argmax.tolist()[0]
+
+        label = self.vocabulary.get_token_from_index(argmax, namespace="labels")
+        prob = round(probs[argmax], 4)
+        return label, prob
+
+
+def main():
+    args = get_args()
+
+    voicemail_detection = VoicemailDetection(model_file=args.voicemail_model_file)
+
+    calling_wav_dir = Path(args.calling_wav_dir)
+    early_media_wav_dir = Path(args.early_media_wav_dir)
+
+    result = list()
+    for filename in tqdm(calling_wav_dir.glob("**/*.wav")):
+        filename = Path(filename)
+        call_id = filename.stem.split("_")[-1]
+
+        # early media
+        early_media_filename = None
+        for fn in early_media_wav_dir.glob("*/*.wav"):
+            basename = fn.stem
+
+            if str(basename).__contains__(call_id):
+                early_media_filename = fn
+                break
+        if early_media_filename is None:
+            continue
+        early_media_label = early_media_filename.parts[-2]
+
+        # read early media signal
+        sample_rate, signal = wavfile.read(early_media_filename)
+        if sample_rate != 8000:
+            raise AssertionError
+        if len(signal) < 1.0 * sample_rate:
+            continue
+        early_media_duration = len(signal) / sample_rate
+
+        # read calling signal
+        sample_rate, signal = wavfile.read(filename)
+        if sample_rate != 8000:
+            raise AssertionError
+        if len(signal) < 1.0 * sample_rate:
+            continue
+        signal = signal[:, 0]
+        calling_duration = len(signal) / sample_rate
+
+        # voicemail
+        calling_label = "non_voicemail"
+        last_prob = 0
+        for i in range(4):
+            begin = i * sample_rate * 2
+            end = begin + sample_rate * 2
+            if end > len(signal):
+                break
+            sub_signal = signal[begin:end]
+
+            label, prob = voicemail_detection.infer(sub_signal)
+            if label == "non_voicemail":
+                prob = 1.0 - prob
+
+            if prob > 0.5:
+                calling_label = "voicemail"
+                break
+            if (prob + last_prob) / 2 > 0.5:
+                calling_label = "voicemail"
+                break
+
+        if early_media_label in ("voicemail",):
+            calling_label = early_media_label
+
+        # print(early_media_label)
+        # print(early_media_duration)
+        # print(calling_label)
+        # print(calling_duration)
+        result.append({
+            "call_id": call_id,
+            "early_media_filename": early_media_filename.as_posix(),
+            "early_media_label": early_media_label,
+            "early_media_duration": early_media_duration,
+            "calling_filename": filename.as_posix(),
+            "calling_label": calling_label,
+            "calling_duration": calling_duration,
+        })
+    result = pd.DataFrame(result)
+    result.to_excel(args.output_file, index=False, encoding="utf-8")
+    return
+
+
+if __name__ == '__main__':
+    main()

examples/early_media_and_voicemail/step_2_analysis_features.py

@@ -0,0 +1,67 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+from collections import Counter
+
+import pandas as pd
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument("--excel_file", default="result.xlsx", type=str)
+
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    df = pd.read_excel(args.excel_file)
+
+    # count
+    labels = df["calling_label"].tolist()
+    counter = Counter(labels)
+    msg = "语音信箱和非语音信箱的数量: \n"
+    msg += "label\tcount\n"
+    for l, c in counter.most_common():
+        row = "{}\t{}\n".format(l, c)
+        msg += row
+    print(msg)
+
+    # voicemail labels
+    labels = df[df["calling_label"] == "voicemail"]["early_media_label"].tolist()
+    voicemail_labels_counter = Counter(labels)
+
+    msg = "语音信箱的早媒体标签特征: \n"
+    msg += "label\tcount\n"
+    for l, c in voicemail_labels_counter.most_common():
+        row = "{}\t{}\n".format(l, c)
+        msg += row
+    print(msg)
+
+    auc = list()
+    voicemail_total = len(df[df["calling_label"] == "voicemail"])
+    non_voicemail_total = len(df[df["calling_label"] == "non_voicemail"])
+    for t in range(60):
+        sub_df = df[df["early_media_duration"] > t]
+        voicemail_count = len(sub_df[sub_df["calling_label"] == "voicemail"])
+        non_voicemail_count = len(sub_df[sub_df["calling_label"] == "non_voicemail"])
+        tpr = voicemail_count / voicemail_total
+        fpr = non_voicemail_count / non_voicemail_total
+        rate = tpr / (fpr + 1e-5)
+        auc.append((t, round(tpr, 4), round(fpr, 4), round(rate, 4)))
+
+    msg = "语音信箱和响铃时长的AUC: \n"
+    msg += "t\ttpr\tfpr\trate\n"
+    for t, tpr, fpr, rate in auc:
+        row = "{}\t{}\t{}\t{}\n".format(t, tpr, fpr, rate)
+        msg += row
+
+    print(msg)
+    return
+
+
+if __name__ == '__main__':
+    main()

examples/make_templates/step_1_wav_classification.py

@@ -24,7 +24,7 @@ from toolbox.cv2.misc import show_image
 from toolbox.python_speech_features.misc import wave2spectrum_image
 
 
-area_code = 60
+area_code = 63
 
 
 def get_args():

examples/make_templates/step_2_wav_split.py

@@ -18,14 +18,14 @@ from tqdm import tqdm
 from project_settings import project_path
 
 
-area_code = 60
+area_code = 66
 
 
 def get_args():
     parser = argparse.ArgumentParser()
     parser.add_argument(
         "--filename",
-        default=(project_path / "data/early_media/60/wav/voice/early_vm_e4543473-eab8-4240-8e9c-d5249a1137b5.wav").as_posix(),
+        default=(project_path / "data/early_media/66/wav/voice/early_vm_0ef2743d-341a-4859-960f-520ed3abda07.wav").as_posix(),
         type=str
     )
     parser.add_argument(

examples/make_templates/step_3_move_by_template.py

@@ -3,6 +3,7 @@
 import argparse
 from collections import defaultdict
 from glob import glob
+import json
 import os
 from pathlib import Path
 import shutil
@@ -18,7 +19,7 @@ from project_settings import project_path
 from toolbox.python_speech_features.misc import wave2spectrum_image
 
 
-area_code = 60
+area_code = 63
 
 
 def get_args():
@@ -170,6 +171,10 @@ def main():
     templates_dir = Path(args.templates_dir)
     wav_dir = Path(args.wav_dir)
 
+    config_json_file = templates_dir / "config.json"
+    with open(config_json_file.as_posix(), "r", encoding="utf-8") as f:
+        config_json = json.load(f)
+
     def wave_to_spectrum(wave: np.ndarray):
         spectrum = wave2spectrum_image(wave=wave, sample_rate=8000)
         spectrum = np.array(spectrum, dtype=np.float32)
@@ -197,15 +202,25 @@ def main():
         if len(matches) == 0:
             continue
 
-        labels = [match['label'] for match in matches]
+        matches_ = list()
+        for match in matches:
+            label = match["label"]
+            matches_.append({
+                **match,
+                "weight": config_json[label]["weight"]
+            })
+        matches_ = list(sorted(matches_, key=lambda x: x["weight"], reverse=True))
+
+        labels = [match["label"] for match in matches_]
         labels_ = [label for label in labels if label not in ("music",)]
 
         if len(set(labels_)) > 1:
             print("超过两个模板类别被匹配，请检测是否匹配正确。")
             print(filename)
-            # for match in matches:
-            #     print(match)
-            continue
+            for match in matches_:
+                print(match)
+            # continue
+            labels_ = labels_[:1]
 
         if len(labels_) == 0:
             label = "music"

examples/media_to_resample_dual_channel.py

@@ -0,0 +1,60 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+import os
+from pathlib import Path
+
+import librosa
+import numpy as np
+from scipy.io import wavfile
+from tqdm import tqdm
+
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--wav_dir",
+        # default=(project_path / "data/early_media/60/wav").as_posix(),
+        default=(project_path / "data/calling/60/wav").as_posix(),
+        type=str
+    )
+    parser.add_argument("--to_sample_rate", default=16000, type=int)
+    parser.add_argument(
+        "--output_dir",
+        # default="./early_media",
+        default="./calling",
+        type=str
+    )
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    wav_dir = Path(args.wav_dir)
+    output_dir = Path(args.output_dir)
+
+    max_value = (2 << 15)
+
+    for filename in tqdm(wav_dir.glob("**/*.wav")):
+        filename = Path(filename)
+
+        signal, sample_rate = librosa.load(filename.as_posix(), sr=args.to_sample_rate)
+        signal = np.array(signal * max_value, dtype=np.int16)
+
+        if signal.ndim != 2:
+            signal2 = np.zeros(shape=signal.shape, dtype=np.int16)
+            signal = np.stack([signal, signal2], axis=-1)
+
+        to_filename = output_dir / filename.relative_to(wav_dir)
+        to_filename.parent.mkdir(parents=True, exist_ok=True)
+        wavfile.write(to_filename.as_posix(), rate=args.to_sample_rate, data=signal)
+    return
+
+
+if __name__ == '__main__':
+    main()

requirements.txt

@@ -16,3 +16,4 @@ gunicorn==20.1.0
 pandas==1.1.5
 xlrd==1.2.0
 openpyxl==3.0.9
+librosa==0.10.1