LiTy: Listen & Type

An Efficient Dictation Training Tool for English Learners.

Train your "human brain speech recognition".

Features

• Support any MP3 file with a transcript.
• Segment audio file automatically into sentences by forced alignment.
• Play segments again and again until you complete the whole sentence.
• Visualize the errors in your answer.
• Record statistics (error rate, speed) for personal reviewing.

Example

• To practice dictation, all you need are a mp3 file and its transcript. Your can find high quality audios and transcripts on Scientific American 60-second science and download them for free.
• When typing sentences, the audio segment will be played repeatly with a 2.5-second pause time. Your can also specify the pause time.
• A visualization of edit distance will be shown after submit your answer. This feature are supported by visedit. We used python-Levenshtein to speed up this module.
• The feedback information will be show and recorded in a log file, including Character Error Rate (CER), word speed per listen iteration.

Setup

Install gentle:

Install submodule gentle:

cd gentle
./install.sh

If you have installed gentle before, skip this step. Just set --gentle_root to the existing directory of gentle when running python main.py.

Install required packages:

• pydub
• pyaudio
• tqdm
• python-Levenshtein

pip install -r requirements.txt

Usage

usage: main.py [-h] --audio AUDIO [--trans TRANS] [--align ALIGN]
               [--begin BEGIN] [--log LOG] [--gentle_root GENTLE_ROOT]
               [--pause_time PAUSE_TIME]

optional arguments:
  -h, --help            show this help message and exit
  --audio AUDIO         The input audio file name.
  --trans TRANS         The input transcript file name.
  --align ALIGN         The input align file name.
  --begin BEGIN         The begining sentence number.
  --log LOG             The log file name.
  --gentle_root GENTLE_ROOT
                        The installed gentle directory.
  --pause_time PAUSE_TIME
                        The pause time between each mp3 replay. (sec)

• When running on an audio file for the first time, gentle will be executed. Alignment file will be save and skip executing gentle when running the second time.
• If you want to begin in at the middle of the whole article, just set --begin [sentence number].
• log file will be in tsv format to record practice time, audio file name, sentence number, CER, word speed/iter, hypothesis, target