mail-tracking-number-voice-recognition-post-processing 0.0.0.1

Library contains functionality for converting given CTC-trained Neural Networks output into tracking number (Russian Post or International mail) format

This package allows you to convert CTC-trained Neural Networks output into mail tracking number in International mail or Russian Post format

Format example for International mail: RO260964943RU
Format example for Russian Post: 1421171600738

Class TrackingNumberRecognizer has method convert(matrix: np.array) -> list, which takes 2d matrix and returns list of possible tracking numbers in decreasing order of probability

Matrix first dimension corresponds to the time moment. Second dimension corresponds to a letter, last index has to correspond to blank (see CTC). Russian letters supported only.

Value is the probability of pronunciation letter at the moment.

Format (International mail or Russian Post) is set in TrackingNumberRecognizer constructor.

You can also set the letters order, the probability of your model making mistake, and the prior probabilities for letters in case of International mail format