- This code uses Python3.
- Run
pip install -r requirements.txt
orpip3 install -r requirements.txt
to install the required dependencies. - Go into
prosogram/
- Run
python3 extract_pitch.py path_to_input_file.wav
- Several prosogram data files will be generated in the location of the input sound file. Our code uses these files.
- The Phrase and Sentence Boundaries will be predicted with their timestamps.
- Speech sample taken as a .wav file
- Prosogram generates the pitch contour, and the prosogram
- Labels given to the pitch contour
- Then trends are extracted from these pitch labels
- Rules written based on these trends try to predict commas, and sentence boundaries.
- If input is given as only one sentence, it can also predict declarative sentences turned into questions.
NOTE: Systematic Evaluation of these predictions have not been done. We have done preliminary evaluations as of now.