Timeline

• Student application work (Mar 25 - Apr 09)

  Get familiar with all the concepts of the project, read documentation and think about possible extensions. Some useful links follow:

  • Sphinx
  • SRILM
  • Spacy
  • scikit-learn
  • Angular
  • Flask

• Community Bonding Period (May 6 - May 26) ✔️

  • Get to know my mentors better and discuss the project more extensively.
  • Implementation of the baseline part of the ASR system based on the default acoustic and language model, that can be found here.
  • Search for speech datasets (recordings along with their transcriptions) and organize them in order to be in Sphinx standard form.
  • Implementation of a domain specific and a merged (specific + default) language model for each dataset.
  • Evaluation of these models in both datasets. Using the default acoustic model and dictionary, the results are here.

• Phase 1 (May 27 - Jun 28) ✔️

  • Extension of the the default dictionary using Phonetisaurus.
  • Adaptation of the acoustic model in all datasets and evaluation.
  • Implementation of a system that extracts the emails from a user's account.
  • Email classification.
  • Email clustering.
  • Create a personal email dataset for evaluation.

• Phase 2 (Jun 29 - Jul 26) ✔️

  • Improve email clustering.
  • Implementation of domain-specific language models.
  • Evaluation of all implemented models in real email speech dataset of my mentor.
  • Implementation of the error detector of the correction system.

• Phase 3 (Jul 27 - Aug 26) ✔️

  • Implementation of the error corrector of the correction system.
  • Implementation of the flask API.
  • Implementation of the angular UI.