Skip to content

Latest commit

 

History

History
35 lines (18 loc) · 2.36 KB

README_lexicon_recipe.md

File metadata and controls

35 lines (18 loc) · 2.36 KB
Raw

Ossian + DNN demo with lexicon

For setup and naive recipe without a lexicon, please see README.md (note that Ossian is implemented in Python 2.7)

DNN-based voice using a recipe with lexicon

Ossian trains voices according to a given 'recipe' -- the recipe specifies a sequence of processes which are applied to an utterance to turn it from text into speech, and is given in a file called $OSSIAN/recipes/<RECIPE>.cfg (where <RECIPE> is the name of a the specific recipe you are using). A recipe called naive_01_nn is described in README.md. Here we will describe the lex_01_nn recipe. If you want to add components to the synthesiser, the best way to start will be to take the file for an existing recipe, copy it to a file with a new name and modify it.

Language resources

For TTS training with a lexicon you need at least a pronunciation dictionary (lexicon), and a phoneset definition. More advanced NLP can involve e.g. part-of-speech tagging, see the English gold standard recipe

Sequitur g2p

Ossian takes care of training a g2p model based on your lexicon, using the Sequitur g2p toolkit. Be sure to add the $OSSIAN/tools/patch/sequitur_compilation.patch to the setup.py of Sequitur before installing, at least if you are using MacOS. If you place Sequitur in $OSSIAN/tools and run the following from the g2p directory: python setup.py install --prefix $OSSIAN/tools , everything will be set up as Ossian expects.

If you use another setup, you have to change the g2p paths in $OSSIAN/scripts/processors/Lexicon.py

lex_01_nn.cfg

This is the configuration for a basic lexicon based training. You might have to adjust this file, some examples:

Change the dictionary parameter for the Lexicon class:

phonetiser = Lexicon('segment_adder', target_nodes="//token", target_attribute='text', child_node_type='segment', \
                            class_attribute='token_class', output_attribute='pronunciation', word_classes = ['word'], \
                            probable_pause_classes = ['punctuation', c.TERMINAL], possible_pause_classes=['space'],\
                            dictionary='my_dictionary_directory', lts_ntrain=1000, lts_gram_length=2)