Multiple stage vector quantization for speech coding

By Biing-Hwang Fred Juang, A. Gray. 1982

Subfields: Efficiency & Compression, Speech & Audio

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Efficiency & Compression

In Speech & Audio

Citations in the corpus

SoundStream: An End-to-End Neural Audio Codec (2021, background): “For this reason, structured vector quantizers [35], [36] (e.g., residual, product, lattice vector quantizers, etc.) have been proposed to obtain a trade-off between computational complexity and coding efﬁciency in traditional codecs.”
High Fidelity Neural Audio Compression (2022, neutral): “Low bitrate parametric speech and audio codecs have long been studied (Atal & Hanauer, 1971; Juang & Gray, 1982), but their quality has been severely limited.”
DelightfulTTS 2: End-to-End Speech Synthesis with Adversarial Vector-Quantized Auto-Encoders (2022, methodology): “Specifically, it consists of a symmetric encoder-decoder network [10, 11, 12, 13] with skip-connections between bottom and top layers, and a multi-stage vector quantizer [14] in the middle as feature bottleneck.”
Introduction to Data Compression (2005, background): “4 Mul t i s tage Vec tor Quant izat ion Multistage vector quantization [147] is an approach that reduces both the encoding complexity and the memory requirements for vector quantization, especially at high rates.”
Autoregressive Image Generation using Residual Quantization (2022, background): “Residual quantization (RQ, also known as stacked quantization) [24,32] iteratively quantizes a vector and its residuals and represents the vector as a stack of codes, which has been used for neural network compression [15, 29, 30].”
MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers (2023, methodology): “These features are quantized as embeddings of a codebook using residual quantization [27, 38], effectively reducing sequence lengths of the mesh representation.”
Polysemous Codes (2016, methodology): “In between PQ and this general formulation, good trade-oﬀs are achieved by residual quantizers [11,31], which are routinely used in the non-exhaustive PQ variant [28] to reduce the quantization loss by encoding the residual error vector instead of the original vector, but also as a coding strategy…”
QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks (2024, methodology): “Using RVQ, we can quantize to 4 bits by rounding with the 2 bit E8P codebook twice.”
Approximate Nearest Neighbor Search by Residual Vector Quantization (2010, background): “Residual vector quantization [19,20] is a common technique to reduce the quantization error with several low complexity quantizers.”