Milestones

Built a voice from existing ARCTIC data (e.g., slt). Worked through individual steps of voice building, starting to understand them in detail. Investigated relative weighting of target cost versus join cost.
Booked your studio training session.

Learned how to make recordings. Started your recordings of ARCTIC A. Decided whether to do automatic script selection, or a manual design.

Completed recording of ARCTIC A script. Fully understood the do_alignment script. Selected your limited domain. For automatic script selection, identified a source of text to select from.

Built a voice from your own ARCTIC A recordings. Implemented your automatic script design algorithm, or manually created your additional script.

Completed all recordings, including your own additional material. Fully understood all the design choices in building a voice. Explored the effect of several design choices (e.g., pitch marking and F0 estimation accuracy). Multiple voices built (e.g., from various subsets of your data or with difference design choices).

Final listening test design started. Decided on method for implementing the test (e.g., online). Made decision about final set of voices required to test your hypotheses. Started building those voices. Identified and fixed one label alignment error.

Listening test designed and implemented. Pilot listening test with one or two listeners completed.

Listening test underway. Continued experimentation to explore design choices.

Optional small follow-on listening test designed. Decided which hypotheses and experiments will feature in your lab report. Selected an evaluation method for each, which will be a listening test in some cases, but not all.

Completed draft write-up.

Confirmed date for 2024: Thursday 11th April