Transform Coded Excitation

Description

Transform Coded Excitation (TCX) is a core coding mode within several 3GPP speech and audio codecs, most notably the Enhanced Voice Services (EVS) codec standardized in Release 12. It is a frequency-domain coding technique used to represent the excitation signal—the residual signal after linear prediction analysis that carries the fine spectral details and noise-like components of the audio. Unlike traditional Code-Excited Linear Prediction (CELP) which operates primarily in the time domain, TCX applies a Modified Discrete Cosine Transform (MDCT) to windowed segments of the excitation signal. This transforms the signal into the frequency domain, where its spectral coefficients can be quantized and encoded efficiently. The TCX frame is typically longer than a CELP frame (e.g., 20ms, 40ms, or 80ms), which allows for better frequency resolution and more efficient coding of stationary signals like music or voiced speech. The quantization of the spectral coefficients uses algebraic vector quantization and employs bandwidth expansion and perceptual noise shaping based on a Linear Predictive Coding (LPC) filter to mask quantization noise. The decoder reconstructs the spectral coefficients, performs an inverse MDCT to get the time-domain excitation, and then filters it through the LPC synthesis filter to produce the output audio. TCX modes are often used in a codec alongside ACELP (Algebraic CELP) modes, with a mode selector choosing the best coding technique for each frame based on signal characteristics, achieving a superior balance between speech quality, music quality, and robustness to transmission errors.

Purpose & Motivation

TCX was created to address the limitations of traditional time-domain CELP codecs, particularly for non-speech audio signals like music and for providing higher quality at lower bitrates. CELP excels at modeling the periodic nature of voiced speech but is less efficient for encoding music, background noise, or unvoiced speech, often resulting in metallic or robotic artifacts. The purpose of TCX is to provide a more generic, transform-based coding tool that delivers consistently high quality for a wider range of audio content, which became critical with the convergence of voice and multimedia services over mobile networks. It solves the problem of inefficient wideband and super-wideband audio coding, enabling high-quality full-band music streaming and voice calls within the same codec framework. The historical motivation was the evolution from narrowband telephony to high-definition voice and beyond, requiring codecs that could support the "voice and audio" service paradigm for LTE and 5G networks. TCX, as part of EVS, directly addressed the competitive need for an codec superior to existing standards like AMR-WB, providing operational flexibility and a better user experience for carriers.

Key Features

• Frequency-domain coding using MDCT transform
• Longer frame lengths for improved coding efficiency of stationary signals
• Integrated perceptual noise shaping based on LPC analysis
• Efficient algebraic vector quantization of spectral coefficients
• Seamless integration with time-domain ACELP modes in a unified codec
• Superior performance for music and mixed content at low to medium bitrates

Evolution Across Releases

Defining Specifications