I am playing with master ( 79e0f6a8fce30556d66eb7afe699b9fb3bab05b6) and tried fine tuning with the 0.1.1 model and got bad results (after training on a specific voice, error rate doesn’t change) . Is this possible or do I need to use the 0.1.1 tag? The code changed between maser and 0.1.1 and the tensorflow required version changed, so im assuming that could be an issue, but I would expect the model to crash during training if this was incompatible.
