/speechtophoneme

This project is developed to create a Deep Learning algorithm able to distinguish English mispronunciations. The idea behind this is to create a mobile app to help people to pronounce English correctly.

Primary LanguagePython

Speech To Phoneme Project

This project is developed to create a Deep Learning algorithm able to distinguish English mispronunciations. The idea behind this is to create a mobile app to help people to pronounce English correctly.

Installation

Requirements

$ pip3 install -r requirements.txt

Installed components:

  • pandas==0.20.3
  • SoundFile==0.10.3.post1
  • matplotlib==2.1.1
  • tensorflow==1.6.0
  • Keras==2.2.5
  • h5py==2.7.1
  • numpy==1.13.3
  • python_speech_features

Import data

  1. Uncompress TIMIT.zip onto ./data directory. The TIMIT/TRAIN directory must be on this subrirectory.
  2. Change directori to data: cd data
  3. Run python3 import_timit_phoneme.py .
  4. Convert wav files into the correct format: python3 prepare_wav.py .
  5. Create mfcc files with python3 create_mfcc.py .

Run with

Sample run:

$ python3 run-train.py --train_files=./data/TIMIT/sample.csv --valid_files=./data/TIMIT/sample.csv --batchsize=2

Bolt run:

$ python3 run-train.py --train_files=./data/TIMIT/timit_phoneme_train.csv --valid_files=./data/TIMIT/timit_phoneme_test.csv --epochs=200 --fc_size=512

Watch results

Run Tensorboard:

$ tensorboard --logdir=/full_path_to_your_logs