Estimation of Interchannel Time Difference in Frequency Subbands Based on Nonuniform Discrete Fourier Transform

Bo Qiu, Yong Xu, Yadong Lu, Jun YangEURASIP J. Audio Speech Music. Process 2008 · DOI

TL;DR

This paper introduces a novel method for estimating interchannel time differences (ICTD) in audio signals by using a nonuniform discrete Fourier transform (NDFT). This approach improves the spatial perception of decoded audio, particularly the width and externalization of the auditory image, without degrading overall sound quality.

What Problem Does This Paper Address?

Spatial audio rendering techniques like Binaural Cue Coding (BCC) rely on side information such as interchannel level difference (ICLD), interchannel time difference (ICTD), and interchannel correlation (ICC) to recreate a sense of space. While ICLD and ICC are well-established, accurately estimating ICTD, especially in lower frequency bands, has been a challenge. Traditional methods using the discrete Fourier transform (DFT) suffer from uniform frequency sampling, which doesn't align well with human auditory perception. This can lead to inaccurate ICTD estimation, potentially degrading audio quality or failing to improve the spatial image. The paper aims to address this by developing a more precise ICTD estimation method that better accounts for perceptual relevance in different frequency bands.

Methods

The core of this paper is the application of the Nonuniform Discrete Fourier Transform (NDFT) for ICTD estimation. Unlike the standard DFT, which samples frequencies uniformly, NDFT allows for flexible placement of frequency bins. This flexibility is leveraged to:

  • Focus on low-frequency bands: The NDFT is configured to allocate more frequency bins to lower frequency ranges, where ICTD is a more critical cue for spatial perception. Conversely, fewer bins are used in higher frequency bands.
  • Subband Analysis: The audio signal is divided into 27 subbands, approximating psychoacoustic critical bands. ICTD is then estimated within each subband.
  • Coherence Factor: A subband coherence factor ($\alpha$) is introduced to determine if ICTD estimation is necessary for a given subband. If the channels are not sufficiently coherent, ICTD is not considered, preventing erroneous estimations.
  • Improved ICTD Calculation: The ICTD is derived from the phase difference between the left and right channels within each subband. The NDFT's ability to concentrate sampling points in relevant frequency regions leads to a more accurate estimation of the phase slope, and consequently, the ICTD.

Datasets

Not discussed in the paper. The paper focuses on a novel algorithm and its evaluation through subjective testing.

Key Insights and Ideas

  • NDFT's Perceptual Advantage: The nonuniform sampling of NDFT is crucial for aligning spectral analysis with human auditory perception, particularly for ICTD estimation. By concentrating spectral resolution in low-frequency bands, NDFT allows for more accurate ICTD estimation where it matters most.
  • ICTD's Role in Spatialization: While sometimes problematic if estimated poorly, ICTD is a vital cue for widening and externalizing auditory images, especially in binaural or HRTF-filtered audio. The paper demonstrates that accurate ICTD estimation significantly enhances these spatial aspects.
  • Coherence as a Prerequisite: The introduction of a coherence factor is a smart way to avoid introducing artifacts. It ensures that ICTD estimation is only performed when there's a genuine time difference to be captured, preventing spurious ICTD values from corrupting the audio.
  • Subjective Evaluation is Key: The paper emphasizes the importance of subjective testing for evaluating spatial audio algorithms. The novel evaluation method for image width provides a more nuanced assessment than traditional objective metrics.

Key Results

  • Improved Auditory Image Width: The NDFT-based ICTD estimation scheme achieved significantly wider and more externalized auditory images compared to the traditional DFT-based BCC scheme and a BCC scheme without ICTD.
  • Superior Overall Quality: Subjective tests showed that the NDFT-based scheme resulted in the highest average scores for overall audio quality among the processed schemes, outperforming both DFT-based ICTD and no-ICTD methods.
  • No Deterioration: Crucially, the NDFT-based ICTD estimation did not deteriorate the sound quality at the decoder compared to the traditional scheme without ICTD estimation.
  • Quantitative Comparison: Average scores for image width were 4.3 for DFT-based ICTD, 3.9 for NDFT-based ICTD, and 3.2 for no-ICTD. For overall quality, the average scores were 3.6 for DFT-based ICTD and 4.1 for NDFT-based ICTD.

Limitations and Future Work

  • Subjective Evaluation Dependence: While the paper introduces a novel subjective evaluation method for image width, the reliance on subjective tests means results can be influenced by listener perception and training.
  • Computational Complexity: While not explicitly detailed, NDFT can sometimes be more computationally intensive than DFT. The paper implies a real-time processing requirement, suggesting that the NDFT implementation is efficient enough.
  • Future Work: The paper suggests that the NDFT-based ICTD estimation is an optimal choice for improving audio quality and image width in BCC. Further exploration into optimizing NDFT configurations for different audio content or perceptual models could be a direction for future research. The paper also mentions that the ICTD is an important cue for headphone playback at frequencies below 1-1.5 kHz, hinting at potential optimizations for specific playback scenarios.

Summary updated

BibTeX
@article{qiu2008estimation,
  author = {Bo Qiu and Yong Xu and Yadong Lu and Jun Yang},
  title = {{Estimation of Interchannel Time Difference in Frequency Subbands Based on Nonuniform Discrete Fourier Transform}},
  journal = {EURASIP J. Audio Speech Music. Process.},
  year = {2008},
  doi = {10.1155/2008/618104},
}
Figure