python api - batch inference possible ?

[phineas-pta]

hello,
thank you for the wonderful work
i have like thousands of text files to generate audio. is it possible to do batch inferencing instead of sequentially generate audio?

[synesthesiam]

This is definitely possible with the PyTorch generator exported directly from the Piper model's checkpoint: https://github.com/rhasspy/piper-sample-generator/blob/2971426a55072f7d22fec416ca7800df8bd23207/piper_sample_generator/__main__.py#L29

See: https://github.com/OHF-Voice/piper1-gpl/blob/main/src/piper/train/export_generator.py

I haven't tried doing batching with the onnx models themselves, but it should be possible by modifying phoneme_ids_to_audio to receive a batch of phoneme ids instead of just a list.

[phineas-pta]

oh yes, i'm only interested with onnx models, thank you for letting me know