Research map / Image generation / Autoregressive and latent generation
NORESQA: A Framework for Speech Quality Assessment using Non-Matching References
Pranay Manocha, Buye Xu, Anurag Kumar
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pre-train speech enhancement model with NORESQA on unpaired data | U-Net CRNN SE, 100% VCTK · VCTK | PESQ: 2.39 -> 2.46 | Improved |
| pre-train SE with NORESQA | U-Net CRNN SE, 100% VCTK · VCTK | STOI: 91.89 -> 93.53 | Improved |
| pre-train SE with NORESQA | U-Net CRNN SE, 33% VCTK · VCTK | PESQ: 2.22 -> 2.28 | Improved |
| use both SNR and SI-SDR heads instead of one head in quantification task (ablation) | NORESQA · MOS correlations | correlation: often more than 30 to 40% improvement | Improved |
| use SI-SDR head only instead of SNR head only (ablation) | NORESQA · MOS correlations | correlation | Improved |
| increase number of NMRs from 1 to 100 (ablation) | NORESQA · MOS correlations | correlation: up to 10 to 15% | Improved |
| use non-matching references instead of paired clean reference (ablation) | NORESQA · VoCo MOS | Pearson: 0.64 -> 0.88 | Improved |
| use non-matching references instead of paired clean reference (ablation) | NORESQA · HiFi-GAN MOS | Pearson: 0.59 -> 0.63 | No clear effect |
| use fixed global NMR set instead of random unpaired NMRs (ablation) | NORESQA · FFTnet MOS | Pearson: 0.46 -> 0.33 | Hurt |
| use NORESQA instead of DNSMOS | 2AFC accuracy · 2AFC HiFi-GAN | accuracy (%): 62.3 -> 81.6 | Improved |
Closest papers by meaning
- Audio Large Language Models Can Be Descriptive Speech Quality Evaluators (ICLR 2025 · similarity 0.90)
- SCOREQ: Speech Quality Assessment with Contrastive Regression (NeurIPS 2024 · similarity 0.88)
- Rethinking Training Targets, Architectures and Data Quality for Universal Speech Enhancement (NeurIPS 2026 · similarity 0.86)
- Self-Supervised Speech Quality Estimation and Enhancement Using Only Clean Speech (ICLR 2024 · similarity 0.85)
- ARECHO: Autoregressive Evaluation via Chain-Based Hypothesis Optimization for Speech Multi-Metric Estimation (NeurIPS 2025 · similarity 0.85)
- AGAV-Rater: Adapting Large Multimodal Model for AI-Generated Audio-Visual Quality Assessment (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
