Sox Can Talk! Training on Rhasspy Piper

Sox can now talk given text! Using Rhasspy Piper, I was able to train Lessac into Sox's voice. It sounds quite good! After multiple false starts due to poor startup documentation, Piper fixed their documentation and I streamlined the process. For this to work, you need Ubuntu with python, and Piper installed, as well as the Lessac voice.

At that point, the commands needed are as follows:

 

python3 -m piper_train.preprocess   --language en-us   --input-dir Sox_Voice/   --output-dir training_dir/   --dataset-format ljspeech   --sample-rate 22050 --single-speaker

 


python3 -m piper_train     --dataset-dir ~/Documents/Rhasspy/python/training_dir/     --devices 1     --batch-size 4     --validation-split 0.0     --num-test-examples 5     --max_epochs 10000     --precision 32 --resume_from_checkpoint ~/Documents/Rhasspy/python/lessacVoice/epoch\=2164-step\=1355540.ckpt

 


head -n5 ~/Documents/Rhasspy/python/training_dir/dataset.jsonl| python3 -m piper_train.infer  --checkpoint ~/Documents/Rhasspy/python/training_dir/lightning_logs/version_0/checkpoints/epoch\=2645-step\=1375742.ckpt --output-dir ~/Documents/Rhasspy/python/output/0

 


python3 piper_train/export_onnx.py  ~/Documents/Rhasspy/python/training_dir/lightning_logs/version_0/checkpoints/epoch\=2645-step\=1375742.ckpt  ~/Documents/Rhasspy/python/output/0/sox.onnx

 

For more info, leave a comment and I'll write more detail, but the instructions are now quite good and easy to follow
 

Comments

You need to log in to post comments