Related Experiment Videos
EigenCodec: Ultra-low bitrate narrowband speech coding via eigen-informed subspace projection and multi-scale
Jingxiang Wang1,2, Ye Li1,2, Peng Zhang1,2
1Key Laboratory of Computing Power Network and Information Security, Ministry of Education, Shandong Computer Science Center (National Supercomputer Center in Jinan), Qilu University of Technology (Shandong Academy of Sciences), Jinan, China.
The Journal of the Acoustical Society of America
|July 21, 2026
Summary
EigenCodec introduces a novel neural speech coding framework for ultra-low bitrates. It achieves efficient narrowband speech reconstruction with enhanced fidelity and reduced computational cost.
Area of Science:
- Signal Processing
- Machine Learning
- Speech Technology
Background:
- Narrowband speech coding at ultra-low bitrates faces challenges due to information loss in compressed latent spaces and computational limits of edge devices.
- Existing methods struggle to balance efficiency and reconstruction quality in highly compressed speech signals.
Purpose of the Study:
- To propose EigenCodec, an efficient neural speech coding framework optimized for ultra-low bitrate narrowband speech.
- To improve representation efficiency and mitigate information loss during dimensionality reduction in speech compression.
Main Methods:
- Introduced a multi-scale depthwise aggregation block combining depthwise convolutions, channel attention, and feed-forward networks for efficient temporal dependency aggregation.
- Developed an eigen-informed subspace projection strategy, using principal component analysis priors to guide quantization and preserve high-variance features.
Main Results:
- EigenCodec demonstrates a competitive balance between computational efficiency and reconstruction fidelity on the LibriSpeech dataset.
- Achieved intelligible and natural-sounding narrowband speech reconstruction at 600 and 800 bps.
Conclusions:
- EigenCodec offers an effective solution for efficient narrowband speech coding at ultra-low bitrates.
- The proposed methods enhance representation efficiency and reduce information loss, outperforming existing baselines in specific scenarios.
Related Concept Videos
01:20Downsampling
When considering a sampled sequence with zero values between sampling instants, one can replace it by taking every N-th value of the sequence. At these integer multiples of N, the original and sampled sequences coincide. This process, known as decimation, involves extracting every N-th sample from a sequence, thereby creating a more efficient sequence.
The Fourier transform of the decimated sequence reveals a combination of scaled and shifted versions of the original spectrum. This...
The Fourier transform of the decimated sequence reveals a combination of scaled and shifted versions of the original spectrum. This...
01:22Upsampling
Managing signal sampling rates is essential in digital signal processing to maintain signal integrity. A decimated signal, characterized by a reduced frequency range due to its lower sampling rate, can be upsampled by inserting zeros between each sample. This upsampling process expands the original spectrum and introduces repeated spectral replicas at intervals dictated by the new Nyquist frequency. To refine this zero-inserted sequence, it is passed through a lowpass filter with a cutoff...
01:26Linear Approximation in Frequency Domain
Linear systems are characterized by two main properties: superposition and homogeneity. Superposition allows the response to multiple inputs to be the sum of the responses to each individual input. Homogeneity ensures that scaling an input by a scalar results in the response being scaled by the same scalar.
In contrast, nonlinear systems do not inherently possess these properties. However, for small deviations around an operating point, a nonlinear system can often be approximated as linear.
In contrast, nonlinear systems do not inherently possess these properties. However, for small deviations around an operating point, a nonlinear system can often be approximated as linear.
01:21Perceiving Loudness, Pitch, and Location
The human brain perceives pitch through two primary mechanisms reflected in place theory and frequency theory. Each mechanism describes how sound waves are interpreted as specific pitches by the brain, offering insights into the intricate processes of auditory perception.
Place theory, or place coding, suggests that different pitches are heard because various sound waves activate specific locations along the cochlea's basilar membrane. The brain determines the pitch of a sound by identifying...
Place theory, or place coding, suggests that different pitches are heard because various sound waves activate specific locations along the cochlea's basilar membrane. The brain determines the pitch of a sound by identifying...
01:17Bandpass Sampling
In signal processing, bandpass sampling is an effective technique for sampling signals that have most of their energy concentrated within a narrow frequency band. This type of signal is known as a bandpass signal. The key principle of bandpass sampling involves sampling the signal at a rate that is greater than twice the signal's bandwidth to prevent aliasing.
A bandpass signal has a spectrum with a lower frequency limit, denoted as ω1, and an upper frequency limit, denoted as ω2. The spectrum...
A bandpass signal has a spectrum with a lower frequency limit, denoted as ω1, and an upper frequency limit, denoted as ω2. The spectrum...
01:10Reconstruction of Signal using Interpolation
Signal processing techniques are essential for accurately converting continuous signals to digital formats and vice versa. When a continuous signal is sampled with a period T, the resulting sampled signal exhibits replicas of the original spectrum in the frequency domain, spaced at intervals equal to the sampling frequency. To handle this sampled signal, a zero-order hold method can be applied, which creates a piecewise constant signal by retaining each sample's value until the next sampling...