Corpora with unit note length as the first token of each tune (e.g. Swedish) will generate poor output if the prime tokens don't precede a set M token with statistically appropriate L token. This handles that.
Generates an appropriate L token for a given M token
i.e. weighted random as per corpus
i.e. none if corpus doesn't have preceding L tokens
i.e. wildcard if M is wildcard
Implemented but disabled: presents an 'ABC Unit' compose option for models that have L: tokens.
Necessary model and processing tweaks for above, including adding unitnotelength property to rnn_tune.
Fixes ordering of header field ordering in folkrnn output processing
Corpora with unit note length as the first token of each tune (e.g. Swedish) will generate poor output if the prime tokens don't precede a set M token with statistically appropriate L token. This handles that.
unitnotelengthproperty tornn_tune.