Tone Trainer Three Thousand

Inspiration

We imagine a world where everyone had equal access to Mandarin learning tools. And then we realized we had to be the change we want to see in the world

What it does

The system allows you to practice pronouncing Chinese words, with examples, and will let you know if you have got the tones correctly. It also includes authentically aesthetic background along with soothing music for your learning pleasure.

How I built it

Blood, sweats and tears

Challenges I ran into

Sleep deprivation.

Accomplishments that I'm proud of

Swag bags.

What I learned

Friendship is Magic.

What's next for Tone Trainer Three Thousand

More data. More joy.

Built With

Submitted to

Hack Cambridge Ternary

Created by

I get data. I make more data from data. I give all data to the machine teacher.

Sỹ Khánh Nguyễn
- Looked into alternative features (MFCC, Filterbanks etc.) and models (CNNs with Keras).

- Extracted additional training data from the THCHS-30 Corpus by aligning the audio with its transcript using SPPAS .

tldr: I get raw data. I align the audio files with their transcripts to make more data. I write code to clean up the data. I send the less terrible data to Khanh. Data.

Charlie Chen
-Designed and constructed the front end to give an immersive experience for users to learn Mandarin

-Worked on the interface between front-end and back-end e.g. calling Matlab or pre-trained model from Python

-Designed methods for user interactions

-Contributed on signal processing theory required in the project

dl541
I extracted the original audio dataset. After the features were extracted, I created the training and validation sets, and created the SVM model which evaluates the audio inputs which are fed into our GUI.

Elena Rastorgueva