Fix CUDA OOM on long scripts by splitting input into chunks

[rachana192837]

Summary
Fixes #157 (CUDA OOM on long scripts).
Long-form scripts (>10 min) caused the 1.5B multi-speaker model to crash due to GPU memory limits.

Changes

• Added long_script_inference.py to split long scripts into smaller chunks.
• Generates audio sequentially and concatenates outputs.
• Supports 1.5B multi-speaker model without crashing.

How to Test

1. Run the 1.5B model with a script longer than 10 minutes using long_script_inference.py.
2. Confirm audio is generated without CUDA OOM errors.
3. Adjust CHUNK_SIZE if GPU memory is low.
   Notes

• This is a workaround to reduce memory usage.

[rachana192837]

@microsoft-github-policy-service agree