Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quartered Chirp Spectral Envelope for Whispered vs Normal Speech Classification

S. Johanan Joysingh, P. Vijayalakshmi, T. Nagarajan

TL;DR

The paper addresses robustly distinguishing whispered from normal speech in front-end classifiers under additive noise. It introduces the quartered chirp spectral envelope (QCSE), which combines the chirp spectrum via the chirp z-transform with the first-quarter spectral envelope, to emphasize pitch-harmonic differences that separate the two speech modes. The QCSE features are learned with a lightweight 1D-CNN and evaluated on the wTIMIT and CHAINS datasets, showing improved accuracy and stability across 0, 5, and 10 dB SNR compared to LFBE+LSTM and QSE+1DCNN baselines, with additional validation that 1D-CNN better captures QCSE patterns than LSTM. This work suggests that chirp-based, spectrally smoothed representations can serve as effective front-end features for robust whispered speech classification, with practical benefits for human-computer interaction and accessibility in noisier environments.

Abstract

Whispered speech as an acceptable form of human-computer interaction is gaining traction. Systems that address multiple modes of speech require a robust front-end speech classifier. Performance of whispered vs normal speech classification drops in the presence of additive white Gaussian noise, since normal speech takes on some of the characteristics of whispered speech. In this work, we propose a new feature named the quartered chirp spectral envelope, a combination of the chirp spectrum and the quartered spectral envelope, to classify whispered and normal speech. The chirp spectrum can be fine-tuned to obtain customized features for a given task, and the quartered spectral envelope has been proven to work especially well for the current task. The feature is trained on a one dimensional convolutional neural network, that captures the trends in the spectral envelope. The proposed system performs better than the state of the art, in the presence of white noise.

Quartered Chirp Spectral Envelope for Whispered vs Normal Speech Classification

TL;DR

The paper addresses robustly distinguishing whispered from normal speech in front-end classifiers under additive noise. It introduces the quartered chirp spectral envelope (QCSE), which combines the chirp spectrum via the chirp z-transform with the first-quarter spectral envelope, to emphasize pitch-harmonic differences that separate the two speech modes. The QCSE features are learned with a lightweight 1D-CNN and evaluated on the wTIMIT and CHAINS datasets, showing improved accuracy and stability across 0, 5, and 10 dB SNR compared to LFBE+LSTM and QSE+1DCNN baselines, with additional validation that 1D-CNN better captures QCSE patterns than LSTM. This work suggests that chirp-based, spectrally smoothed representations can serve as effective front-end features for robust whispered speech classification, with practical benefits for human-computer interaction and accessibility in noisier environments.

Abstract

Whispered speech as an acceptable form of human-computer interaction is gaining traction. Systems that address multiple modes of speech require a robust front-end speech classifier. Performance of whispered vs normal speech classification drops in the presence of additive white Gaussian noise, since normal speech takes on some of the characteristics of whispered speech. In this work, we propose a new feature named the quartered chirp spectral envelope, a combination of the chirp spectrum and the quartered spectral envelope, to classify whispered and normal speech. The chirp spectrum can be fine-tuned to obtain customized features for a given task, and the quartered spectral envelope has been proven to work especially well for the current task. The feature is trained on a one dimensional convolutional neural network, that captures the trends in the spectral envelope. The proposed system performs better than the state of the art, in the presence of white noise.
Paper Structure (9 sections, 3 equations, 4 figures, 2 tables)

This paper contains 9 sections, 3 equations, 4 figures, 2 tables.

Table of Contents

  1. Abstract
  2. Introduction
  3. Proposed Feature
  4. Experimental Setup and Results
  5. Network Architecture
  6. Datasets
  7. Evaluation
  8. Results and Discussion
  9. Conclusion

Figures (4)

  • Figure 1: Difference between the spectral envelope of normal and whispered speech
  • Figure 2: The QSE and QCSE (computed at $r=1.010$) of Normal and Whispered speech.
  • Figure 3: The QSE and QCSE (computed at $r=1.010$) of Normal and Whispered speech that is corrupted by noise at 5dB SNR.
  • Figure 4: 1D-CNN network architecture used in the current work