This research introduces a novel approach using auditory large language models to evaluate speech quality across multiple dimensions, replacing traditional single-task models.

• Enables comprehensive assessment of speech quality metrics (MOS, speaker similarity, A/B testing)
• Leverages task-specific prompts to finetune auditory LLMs for quality prediction
• Demonstrates superior performance compared to conventional speech evaluation models
• Provides a unified solution that can handle diverse speech quality assessment tasks

For engineering teams, this breakthrough offers a more efficient and accurate method for speech quality evaluation in audio processing systems, with potential applications in voice assistants, call centers, and audio production tools.

Enabling Auditory Large Language Models for Automatic Speech Quality Evaluation