Skip to content

Latest commit

 

History

History
7 lines (5 loc) · 563 Bytes

README.md

File metadata and controls

7 lines (5 loc) · 563 Bytes

Mandarin-Tone-Recognition

Two Jupyter notebooks detailing an attempt to classify the tones of a Mandarin Chinese utterance based on the audio spectrogram. Part 1 uses a manual extraction of fetures from extracted tone lines; Part 2 applies the full spectrogram as the input of a convolutional neural network; Part 3 uses a recurrent neural network to encode the variable length input spectrogram.

The model obtained in Part 3 is used for tone clasificaiton in the related demo Android application.