Download MSGD_dataset_final.json and place it in this directory:
data/MSGD_dataset_final.json
The dataset contains 500 multi-phase sales dialogues with the following structure:
- Chitchat → Transition → Task-oriented phases
- Intents: FindAttraction, FindRestaurants, FindMovie, LookupMusic, SearchHotel, FindEvents, GetTransportation, SearchFlights
After running scripts/1_prepare_user_audio.py, user audio files will be placed here:
data/user_audio/
merge_0042_turn00.wav # Dialogue 0042, user turn 0 (even = user turns)
merge_0042_turn02.wav # Dialogue 0042, user turn 2
merge_0042_turn04.wav # ...
...
To use voice cloning with Sesame CSM, place voice profile samples here:
data/voice_profiles/
user/
sample_01.wav
sample_02.wav
...