generate_voices_bin.py generates the Sherpa-ONNX voice table from the
official Kokoro-82M/voices directory. The generated v1.0 table contains 54
voices, including the Spanish em_santa voice at speaker ID 53. Existing
speaker IDs 0 through 52 remain unchanged.
The generator validates the style embedding shape (510 x 1 x 256) and
float32 dtype before writing the binary file. Run it from this directory
after downloading the Kokoro-82M repository:
python3 generate_voices_bin.py
The corresponding model metadata must be generated with add_meta_data.py,
which derives n_speakers, speaker_names, id2speaker, and speaker2id
from the same mapping.
