GPT-4: Researchers have adapted OpenAI’s Whisper model, an automatic speech recognition system, to perform unseen tasks using simple prompts. The study, called PromptingWhisper, investigates the zero-shot task generalization abilities of the model and focuses on three specific tasks: audio-visual speech recognition, code-switched speech recognition, and speech translation. By designing task-specific prompts, the researchers significantly improved performance across the tasks, with gains ranging from 10% to 45%. The study highlights Whisper’s robustness to different prompts, accent biases, and multilingual understanding.
Read more at MarkTechPost…