open-loop pitch search

Description

Open-loop pitch search is a fundamental signal processing technique used in speech codecs, particularly in the Adaptive Multi-Rate (AMR) and Adaptive Multi-Rate Wideband (AMR-WB) codecs standardized by 3GPP. Its primary function is to estimate the pitch period (also called lag) of a speech signal in an efficient manner. Pitch, in the context of speech, refers to the fundamental frequency of voiced sounds, which corresponds to the vibration rate of the vocal cords. Accurate pitch estimation is crucial for achieving high compression efficiency in speech codecs, as it allows the codec to exploit the periodicity of voiced speech.

The open-loop pitch search operates on the weighted speech signal, which is the input speech after filtering and perceptual weighting. The algorithm computes a correlation or a similar metric between the current frame and a range of delayed versions of the signal to find the lag that maximizes the similarity. This process is called "open-loop" because it does not involve any feedback from the subsequent quantization stages; it is a direct estimation from the input signal. The result is a coarse pitch estimate that is then used to define a small search window around which the more precise closed-loop pitch search is performed. This two-stage approach significantly reduces the computational complexity compared to performing a full closed-loop search over the entire pitch range.

In the AMR-WB codec, the open-loop pitch search is performed in every other subframe (i.e., once per frame, since a frame consists of four subframes). The search is typically conducted over a range of lags corresponding to pitch frequencies from about 50 Hz to 400 Hz. The algorithm may use decimated signals or simplified correlation measures to further reduce complexity. The output of the open-loop search is a set of candidate lags that are passed to the closed-loop search, which then refines the estimate using analysis-by-synthesis techniques to minimize the perceptually weighted error.

The open-loop pitch search is also employed in other codecs such as AMR-NB, GSM-EFR, and various ITU-T codecs. Its design involves trade-offs between accuracy and complexity. A more accurate open-loop estimate can reduce the burden on the closed-loop search, but it may require more computations. Conversely, a simpler open-loop search may lead to a larger closed-loop search range, increasing overall complexity. The 3GPP specifications (e.g., TS 26.090, 26.190, 26.290) provide detailed algorithms for the open-loop pitch search, ensuring interoperability and consistent performance across implementations.

In summary, the open-loop pitch search is a critical component of modern speech codecs, enabling efficient and accurate pitch estimation. It leverages the periodicity of speech to reduce the search space for the more precise closed-loop analysis, thereby achieving a balance between quality and computational cost. Its implementation in AMR-WB and other codecs has contributed to the widespread adoption of these codecs in mobile telephony and beyond.

Purpose & Motivation

The open-loop pitch search exists to solve the problem of efficiently estimating the pitch period in speech coding. Pitch estimation is essential for exploiting the redundancy in voiced speech, but performing a full, high-resolution search for the optimal pitch lag in every subframe is computationally expensive. The open-loop search provides a coarse estimate that narrows the search range for the subsequent closed-loop search, thereby reducing the overall complexity of the codec. This is particularly important for mobile devices with limited processing power and battery life.

Before the advent of open-loop pitch search, early speech codecs either used simpler but less accurate pitch estimation methods or performed exhaustive searches that were impractical for real-time applications. The open-loop approach was developed to strike a balance between accuracy and complexity. By operating directly on the weighted speech signal and using efficient correlation techniques, it provides a reliable estimate of the pitch lag without the need for iterative analysis-by-synthesis. This makes it suitable for real-time implementations in mobile handsets and network equipment.

The motivation for including open-loop pitch search in 3GPP codecs like AMR-WB was to achieve high speech quality at low bit rates while maintaining low computational complexity. The AMR-WB codec, standardized in Release 8, is used in HD voice services and requires efficient pitch analysis to deliver superior voice quality. The open-loop search is a key enabler for this, as it allows the codec to focus its computational resources on the most promising pitch lags. Its continued presence in subsequent releases underscores its effectiveness and the ongoing need for efficient speech coding in mobile networks.