Speech frame

Description

In 3GPP speech codec specifications, a speech frame is a fundamental unit of encoded voice data. It is a traffic frame that has been classified into one of three categories: SPEECH_GOOD, SPEECH_BAD, or SPEECH_LOST. These classifications are used in the context of speech codec testing, performance evaluation, and error concealment. A SPEECH_GOOD frame indicates that the frame contains valid speech data that can be decoded without errors. A SPEECH_BAD frame indicates that the frame contains errors or is corrupted, but some information may still be usable. A SPEECH_LOST frame indicates that the frame is completely missing or unusable, and error concealment techniques must be applied to mitigate the impact on voice quality. Speech frames are generated by speech codecs such as AMR, AMR-WB, EVS, and others. Each codec operates on a specific frame length, typically 20 ms, and produces a certain number of bits per frame depending on the codec mode. The classification of speech frames is often performed during subjective or objective speech quality testing, where the reference and degraded signals are compared frame by frame. In network simulations and drive tests, speech frames are classified to assess the impact of transmission errors, packet loss, and jitter on voice quality. The classification also plays a role in the receiver's error concealment algorithms: when a frame is lost or bad, the decoder may extrapolate parameters from previous good frames to generate a replacement frame, reducing the perceived distortion. The terms SPEECH_GOOD, SPEECH_BAD, and SPEECH_LOST are defined in specifications such as 26.092, 26.093, 26.192, 26.193, and various 46-series specs for GERAN. These specifications describe the mapping between frame classification and quality metrics like Mean Opinion Score (MOS) or Frame Erasure Rate (FER). Understanding speech frame classification is essential for optimizing voice services over cellular networks, especially in challenging radio conditions where packet loss and errors are common. It enables engineers to evaluate codec robustness, tune error concealment, and ensure acceptable voice quality for users.

Purpose & Motivation

The concept of a speech frame and its classification exists to provide a standardized way to evaluate and manage voice quality in digital cellular networks. Before the advent of digital speech codecs, voice quality was primarily assessed using analog metrics. With digital transmission, voice is encoded into frames, and errors or losses can occur due to radio link impairments. The classification into SPEECH_GOOD, SPEECH_BAD, and SPEECH_LOST allows network operators and equipment vendors to quantify the impact of transmission errors on speech intelligibility and naturalness. It also enables the development of error concealment algorithms that can mask the effects of lost or corrupted frames, improving the user experience. The classification is used in conformance testing to ensure that codecs and transmission systems meet minimum performance requirements. Furthermore, it facilitates objective speech quality assessment, where algorithms like PESQ or POLQA compare the degraded speech signal to a reference, frame by frame, to produce a quality score. The need for such classification became apparent as cellular networks evolved from analog to digital, and as packet-switched voice (VoIP, VoLTE) introduced new challenges like packet loss and jitter. By defining clear categories for speech frames, 3GPP provided a common language for engineers to discuss and address voice quality issues, leading to improved codec designs and network optimization strategies.