Language Specific Decoding With Prompt Engineering

via “language-specific decoding with prompt engineering”

Robust Speech Recognition via Large-Scale Weak Supervision

Unique: Integrates prompt conditioning directly into beam search without requiring fine-tuning, enabling rapid iteration on vocabulary biasing without retraining; uses model's existing language understanding rather than external vocabulary lists.

vs others: Faster than fine-tuning for vocabulary adaptation; less effective than domain-specific models but requires no labeled data or training infrastructure.