Rerun for IEMOCAP - Githubissues

SpectData / MONAH

Multi-Modal Narratives for Humans

MIT License

3 stars 0 forks source link

Rerun for IEMOCAP #6

Closed joshkyh closed 3 years ago

joshkyh commented 4 years ago

Ability to bring your own transcript. Analyst has to insert into the table that defines talkturns. (Not word-level timings)
Target is to fully reproduce iemoca.Narrrative_Fine

joshkyh commented 3 years ago

BYO transcript is named weaved talkturn .csv currently placed in stage 2.

Demo failed after Stage 1. To debug offline and demo complete run on Tues. Build the coarse config file into the existing config. Please ensure that it is easy to the navigate the many options.

mfmakahiya commented 3 years ago

Fixed the bug in re-running iemocap.

joshkyh commented 3 years ago

(1) Create Narrative_Fine.csv so that it is easy for the easy to know where to find their fine narratives.

(2) Implement assertion called stage_3/test_main.py that asserts the first 10 talkturn is exactly as IEMOCAP. Once this is done, we will have to swap it to Joshua/Marriane demo. Because users can't be expected have IEMOCAP videos.

(3) Do (2) for vpa, va, vp, v.

joshkyh commented 3 years ago

(1) There are no test case for va, please feel free to delete the test case for va.

joshkyh commented 3 years ago

Done with 309312dd0cf07617a419b02c3aef214c2d08126e