FiLM-Based Speaker Conditioning of a SpeechLLM for Pathological Speech Recognition
Researchers have developed a new method for improving automatic speech recognition (ASR) for pathological speech using a technique called Feature-wise Linear Modulation (FiLM). This approach injects speaker-specific information into a frozen ASR encoder, allowing it to adapt to individual speakers without altering the base model's weights. The method was benchmarked against existing adaptation strategies on Spanish and English pathological speech datasets, showing competitive performance while maintaining the ability to answer speech-related questions. AI
IMPACT This research could lead to more accurate speech recognition for individuals with neurological conditions, improving accessibility and communication tools.