Hangover period

Description

The hangover period is a concept used in speech codecs within 3GPP systems, particularly in the context of Discontinuous Transmission (DTX) and Comfort Noise Generation (CNG). It is defined in specifications such as 26.093, 26.193, 46.041, and 46.081. During a typical voice call, a speech encoder uses Voice Activity Detection (VAD) to determine whether the current frame contains active speech or silence. When silence is detected, the encoder can switch to a lower bit rate or stop transmitting altogether to save bandwidth and power. However, an abrupt transition from speech to silence can cause artifacts and make it difficult for the decoder to generate appropriate comfort noise. The hangover period addresses this by extending the active speech transmission for a few frames after the VAD has indicated silence.

Technically, the hangover period consists of a number of frames added at the end of a speech burst. During these frames, the VAD flag is set to "0" (indicating no active speech), but the TX_TYPE is set to "SPEECH_GOOD". This means that the encoder continues to encode and transmit the frames as if they were active speech, even though the VAD has already detected silence. The purpose is to provide the encoder with an extra window of samples to analyze and derive the Comfort Noise parameters. Comfort noise is a low-level noise generated at the receiver to fill the silence periods, making the call sound more natural and avoiding the impression that the line has gone dead. By using the hangover period, the encoder can better estimate the background noise characteristics, leading to more accurate comfort noise generation.

The length of the hangover period is typically defined in the codec specification. For example, in the Adaptive Multi-Rate (AMR) codec, the hangover period may consist of a certain number of frames (e.g., 7 frames for AMR). The exact number depends on the codec and its configuration. The hangover period is part of the overall DTX/CNG mechanism, which includes functions such as VAD, comfort noise parameter estimation, and comfort noise generation. The encoder uses the frames in the hangover period to update its noise estimation, and then it transitions to transmitting Silence Insertion Descriptor (SID) frames that contain the comfort noise parameters. The decoder uses these parameters to generate comfort noise during silence periods.

The hangover period is crucial for maintaining voice quality during DTX operation. Without it, the transition from speech to silence might be too abrupt, causing the comfort noise to be inaccurate or the background noise to change suddenly. This could lead to user dissatisfaction. The hangover period ensures a smooth transition by allowing the encoder to gather enough data to characterize the background noise accurately. It also helps in avoiding clipping of the end of speech bursts, which could occur if the encoder switched to silence mode too early.

In 3GPP specifications, the hangover period is defined in the context of specific codecs. For example, in the AMR and AMR-WB codecs, the hangover period is specified as part of the DTX functionality. The specifications 26.093 and 26.193 cover the AMR and AMR-WB codecs, respectively, while 46.041 and 46.081 cover the half-rate and full-rate codecs in GERAN. The term has remained consistent across releases, indicating its fundamental role in speech coding. The hangover period is an essential component of efficient voice transmission over cellular networks, balancing bandwidth savings with voice quality.

Purpose & Motivation

The hangover period exists to improve the quality of comfort noise generation during Discontinuous Transmission (DTX) in speech codecs. DTX is used to save bandwidth and power by not transmitting during silence periods. However, when the encoder stops transmitting speech frames, the decoder must generate comfort noise to fill the silence. To generate accurate comfort noise, the encoder needs to analyze the background noise characteristics. The hangover period provides a window of frames after the end of a speech burst where the encoder can continue to analyze the background noise and derive the comfort noise parameters. Without this period, the comfort noise might be inaccurate, leading to unnatural sounding silence or abrupt changes in background noise.

The motivation for the hangover period comes from the need to balance efficiency and quality. Early speech codecs without DTX transmitted continuously, wasting resources during silence. When DTX was introduced, it saved resources but sometimes degraded quality due to poor comfort noise. The hangover period was introduced as a solution to allow the encoder to gather sufficient data for comfort noise estimation without sacrificing the benefits of DTX. It ensures that the transition from speech to silence is smooth and that the comfort noise matches the actual background noise as closely as possible.

Historically, the concept of hangover period has been part of speech coding standards since the early days of digital cellular. It is included in codecs like GSM HR, FR, EFR, AMR, and AMR-WB. The term itself is descriptive: it refers to the period where the encoder "hangs over" from speech to silence. Over releases, the definition has remained largely unchanged, indicating its fundamental importance. The hangover period is a small but critical detail in the overall design of efficient and high-quality voice communication systems.