Skip to content

Latest commit

 

History

History
98 lines (64 loc) · 2.51 KB

recipe_project.md

File metadata and controls

98 lines (64 loc) · 2.51 KB

Required tools

git clone -b project https://github.com/CSTR-Edinburgh/ophelia.git
git clone https://github.com/AvashnaGovender/Merlyn.git
git clone https://github.com/AvashnaGovender/Tacotron.git

DCTTS + WaveRNN

See recipe here.

Attention experiments

Obtaining forced alignment labels:

Step 5a - Run forced alignment in:
https://github.com/AvashnaGovender/Tacotron/blob/master/PAG_recipe.md

FA as target

Convert forced alignment labels to the forced alignment matrix:

Step 6 - Get durations and create guides:
https://github.com/AvashnaGovender/Tacotron/blob/master/PAG_recipe.md

To use FA as target in DCTTS see config file:
fa_as_target.cfg

FA as guides

Create attention guides from forced alignment matrix

cd ophelia/
python convert_alignment_to_guide.py fa_matrix.npy fa_guide.npy 

To use FA as guide in DCTTS see config file:
fa_as_guide.cfg

FA as attention

Add phone level duration to transcript.csv using forced alignment matrix

cd ophelia/
python add_duration_to_transcript.py fa_matrix_dir transcript_file new_transcript_file

To use FA as attention in DCTTS see config file:
fa_as_attention.cfg

Text Encoder experiments

Labels -/+ TE

Convert state labels to 416 normalised label features (needs state labels and question file)

cd Merlyn/
python scripts/prepare_inputs.py

To use Labels-TE in DCTTS see config file:
labels_minus_te.cfg

To use Labels+TE in DCTTS see config file:
labels_plus_te.cfg

To use C-Labels+TE in DCTTS see config file:
c-labels_plus_te.cfg

PE&Labels + TE

Create new transcription file with phoneme sequence from labels by replace phone sequence of transcript file with phone sequence from HTS style labels

cd ophelia/
./labels2tra.sh labels_dir transcript_file new_transcript_file

To use PE&Labels+TE set MerlinTextEncWithPhoneEmbedding to True in the config file.

Gross error detection experiments

To calculate CDP, Ain and Aout:

cd ophelia/
python calculate_CDP_Ain_Aout.py attention_matrix.npy

FIA experiments

To generate without FIA (forcibly incremental attention) set turn_off_monotonic_for_synthesis to True in the config file.

Tacotron experiments

See readme in Tacotron repository: https://github.com/AvashnaGovender/Tacotron