Research map / Architectures and efficiency / Quantization
Two-dimensional quantization for geometry-aware audio coding
Tal Shuster, Eliya Nachmani
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 |
|---|---|---|---|
| use rhombic grid instead of rectangle grid in Q2D2 (ablation) | Q2D2 · LibriTTS test-clean | PESQ: 2.2909 -> 2.3995 | Improved |
| use rhombic grid instead of hexagonal grid in Q2D2 (ablation) | Q2D2 · LibriTTS test-clean | PESQ: 2.2862 -> 2.3995 | Improved |
| use 6 latent dimensions instead of 8 (lower resolution) at 1 kbps (ablation) | Q2D2 · LibriTTS test-clean | PESQ: 2.092 -> 2.3995 | Improved |
| use 6 latent dimensions instead of 4 (higher resolution) at 1 kbps (ablation) | Q2D2 · LibriTTS test-clean | PESQ: 2.018 -> 2.3995 | Improved |
| increase grid levels from [7]x6 to [11]x6 (ablation) | Q2D2 · LibriTTS test-clean | PESQ: 2.3995 -> 4.2832 PESQ; codebook utilization 92.18% -> 72.11% | Mixed |
| replace VQ (WavTokenizer) quantizer with Q2D2 in same framework (ablation) | WavTokenizer framework · LibriTTS test-clean (LibriTTS-trained) | PESQ: 2.4655 -> 2.5021 | Improved |
| replace FSQ with Q2D2 in same framework (ablation) | WavTokenizer framework · LibriTTS test-clean (LibriTTS-trained) | PESQ: 2.3873 -> 2.5021 | Improved |
| Q2D2 at 1 kbps vs WavTokenizer on LJSpeech (out-of-domain) | Q2D2 1kbps · LJSpeech | UTMOS: 4.2580 -> 3.9715 | Hurt |
| use Q2D2 single quantizer (3.3 kbps) instead of multi-quantizer DAC 9 kbps | codec · MUSHRA | MUSHRA score: 92.64 -> 98.05 | Improved |
| use Q2D2 tokens instead of WavTokenizer tokens for TTS | ParlerTTS 600M · LibriTTS TTS | CMOS-Q: -0.22 -> -0.11 | Improved |
Closest papers by meaning
- WavTokenizer: an Efficient Acoustic Discrete Codec Tokenizer for Audio Language Modeling (ICLR 2025 · similarity 0.86)
- Approaching Rate-Distortion Limits in Neural Compression with Lattice Transform Coding (ICLR 2025 · similarity 0.84)
- Scaling Transformers for Low-Bitrate High-Quality Speech Coding (ICLR 2025 · similarity 0.84)
- Qinco2: Vector Compression and Search with Improved Implicit Neural Codebooks (ICLR 2025 · similarity 0.84)
- SECodec: Structural Entropy-based Compressive Speech Representation Codec for Speech Language Models (AAAI 2025 · similarity 0.84)
- ALMTokenizer: A Low-bitrate and Semantic-rich Audio Codec Tokenizer for Audio Language Modeling (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.
