Telecom Voice Classification

Contents

• televid.py: Provide the basic model for identification of single wave file.
• run_through.py: Only for automatically test every mp3 and wave file in ./test_audio folder.
• run_televid_example.py: The example for using the Televid model in televid.py.
• /golden_wav: The folder contains wave files to generate golden patterns for matching. The golden_ptns.pkl file in it is to speed up the loading (if exists).
• /python_speech_features: The package for MFCC feature.

Categories

• In Busy
  • Original Texts
    • 您所撥的電話忙線中，請稍後再撥。 The number you have dial is busy. Please try again later.
  • Keywords
    • 忙線中
• No Response
  • Original Texts
    • 您所撥的電話無人回應。
    • 您所撥的電話無法接聽，請稍後再撥。The number you has dial is not available. Please try again later.
  • Keywords
    • 回應
    • 接聽
• Voice Mail
  • Original Texts
    • 您的電話將轉接到語音信箱，嘟聲後開始計費，如不留言請掛斷。快速留言嘟聲後請按＃字鍵這是09XX-XXXXXX的信箱，嗶聲後請留言。
    • 轉接語音信箱，嘟聲後開始計費，如不留言請掛斷。快速留言嘟聲後請按 * 字鍵。您已進入09XX-XXXXXX的信箱，嗶聲後請留言。
    • 嘟聲後開始計費，如不留言請掛斷。快速留言嘟聲後請按 * 字鍵。您已進入09XX-XXXXXX的信箱，嗶聲後請留言。
    • 您的電話將轉接到語音信箱，嘟聲後開始計費，如不留言請掛斷。快速留言嘟聲後請按一次 * 字鍵。
  • Keywords
    • 語音信箱
    • 嘟聲後

Spec of dataset.pkl

The file is generated by calling save_mfcc_training_dataset() of TestTelevid object.

File Format

Is the pickle, Python object serialization, object. You should use pickle.loads() to open it.

Content Format

A list contains multiple tuples, and being one data input, each tuple contains two variables: difference indices dictionary (input) and the final outcome (desire output). The following is the example structure.

[
    (
        {'in_busy.wav': 285.38331120979802,
         'no_response_A.wav': 3430.9011939973934,
         'no_response_B.wav': 2380.5107159615013,
         'voice_mail_A_1.wav': 3118.0131102683249,
         'voice_mail_A_2.wav': 2543.842005054099,
         'voice_mail_B.wav': 3217.5192698396595,
         'voice_mail_C.wav': 3176.1896581534188,
         'voice_mail_D_1.wav': 2749.3900560206898,
         'voice_mail_D_2.wav': 2634.9694942389929},
     'inbusy'),
    (
        {'in_busy.wav': 2306.9580628421327,
         'no_response_A.wav':3727.4782066043199,
         'no_response_B.wav': 2987.7147191635795,
         'voice_mail_A_1.wav': 2345.7121554450459,
         'voice_mail_A_2.wav': 2477.9789650889234,
         'voice_mail_B.wav': 328.95530907943493,
         'voice_mail_C.wav': 332.53955702325237,
         'voice_mail_D_1.wav': 2714.4307011467545,
         'voice_mail_D_2.wav': 1705.1556381267567},
     'voicemail')]

Trained Module Input Requirements

The input of the trained module must be a dict() and then return the classified result.

Dependencies

SciPy>=1.1.0

pip install scipy

NumPy>=1.15.0

pip install numpy

FFmpeg-Python>=0.1.16

pip install ffmpeg-python

FFmpeg (for the FFmpeg-Python package)

apt-get install ffmpeg