This project runs the backend flask server, where it accepts an audio file, carries out analysis on the audio file, and returns the corresponding output results in a response.
- Azure Account
- PostgreSQL
- To rerun the model using another set of data, change the string pointing to the location of the raw data source.
- Depending on the data source, certain preprocessing steps can be ommitted.
- Remove silence from the raw audio file.
- Remove "Ellie's" voice from the audio file.
- Segment the audio file into 10 second intervals using pyAudio.
- Create spectograms for each interval.
- Audio is being passed to Amazon S3 (storage), Amazon Transcribe (speech-to-text), audio spectogram model.
- The text that is returned is being passed to Amazon Comprehend (text based sentiment analysis) and local NLP module (psychological absolutist words).
- The audio spectogram model would return a 1/0 classifier score (depressed / normal).
- The scores returned from all 3 modules would be weighted and a final output score would be returned.
- The results from each module would be saved into postgresql database. (Change URL / endpoint for cloud database).
- Tensorflow 1.13.0
- scikit-learn
- OpenCV
1.0
Alvin Tan Jian Jia