Batch Synthesis

Generate many WAVs in a single call, sharing one loaded model.

Generate many WAVs in one call, sharing a single loaded model.

from strands_omnivoice import omnivoice_batch

items = [
    {"id": "en", "text": "Hello world", "instruct": "female, british accent"},
    {"id": "fr", "text": "Bonjour le monde", "language": "French"},
    {"id": "ja", "text": "こんにちは", "language": "Japanese"},
]

omnivoice_batch(items=items, output_dir="/tmp/batch")

Output: /tmp/batch/en.wav, /tmp/batch/fr.wav, /tmp/batch/ja.wav.

Item Schema

Key	Required	Notes
id	✅	Output filename stem (<id>.wav)
text	✅	Text to synthesize
ref_audio	❌	Triggers clone mode
ref_text	❌	Optional reference transcript
instruct	❌	Triggers design mode
language	❌	Language hint (English, en, ...)
duration	❌	Fixed length in seconds
speed	❌	Speech speed factor

If neither ref_audio nor instruct is set, the item runs in auto mode.

JSON String / JSONL Path

For convenience, items can be:

• a Python list (above)
• a JSON-encoded string of a list
• a path to a .jsonl file (one item per line)

omnivoice_batch(items="/path/to/test.jsonl", output_dir="/tmp/out")

Multi-GPU Scaling

For thousands of items across multiple GPUs, use the upstream CLI directly:

omnivoice-infer-batch --model k2-fsa/OmniVoice --test_list test.jsonl --res_dir results/

omnivoice_batch is sequential — best for small/medium jobs.