Techniques for improving the performance of CELP (code excited linear prediction)-type speech coders while maintaining reasonable computational complexity are explored. A harmonic noise weighting function, which enhances the perceptual quality of the processed speech, is introduced. The combination of harmonic noise weighting and subsample pitch lag resolution significantly improves the coder performance for voiced speech. Strategies for reducing the speech coder's data rate, while maintaining speech quality, are presented. These include a method for efficient encoding of the long-term predictor lags, utilization of multiple gain vector quantizers, and a multimode definition of the speech coder frame. A 5.9-kb/s VSELP speech coder that incorporates these features is described. Complexity reduction techniques which allow the coder to be implemented using a single fixed-point DSP (digital signal processor) are discussed. >

Techniques for improving the performance of CELP-type speech coders

Techniques for improving the performance of CELP (code excited linear prediction) type speech coders while maintaining reasonable computational complexity are explored. A harmonic noise weighting function which enhances the perceptual quality of the processed speech is introduced. The combination of harmonic noise weighting and subsample resolution pitch significantly improves the coder performance for voiced speech. A 6.9 kb/s VSELP speech coder which incorporates subsample resolution pitch and harmonic noise weighting is described. Complexity reduction techniques are discussed which allow the coder to be implemented using a single fixed point digital signal processor. >

Techniques for improving the performance of CELP type speech coders

A method of determining a pitch period of an audio signal using a correlation-based signal derived from the audio signal. The correlation-based signal includes known peaks each corresponding to a respective one of known time lags. The known peaks includes a global maximum peak. The method comprises: (a) determining if a candidate peak among the local peaks exceeds a peak threshold; (b) determining if a candidate time lag corresponding to the candidate peak is within a predetermined range of at least one integer sub-multiple of the time lag corresponding to the global maximum peak; and (c) setting the pitch period equal to the candidate time lag when the determinations of both steps (a) and (b) are true.

Pitch extraction methods and systems for speech coding using multiple time lag extraction

A method of searching for an interpolated peak of a Normalized Correlation Square (NCS) signal derived from an audio signal, comprises: producing quadratically interpolated correlation (QIC) signal values at interpolated time lags; squaring each of the QIC signal values to produce square QIC signal values; producing an individual interpolated energy signal value corresponding to each of the square QIC signal values, wherein ratios of the square QIC signal values to their corresponding interpolated energy values represent interpolated NCS signal values; and selecting, as the interpolated peak, a largest interpolated NCS signal value among the interpolated NCS signal values without evaluating the ratios.

Pitch extraction methods and systems for speech coding using interpolation techniques

Four voice packet reconstruction methods used for speech coded by code excited linear prediction (CELP)-type speech coders are described. In the first method, the authors generalize the waveform substitution technique originally developed for the PCM coded speech to the CELP speech coding. In the second method, a priority level is assigned to each speech frame to protect against those perceptually important and hard-to-reconstruct speech frames being lost. The third and fourth methods both split the information bits in a frame into two groups of different levels of importance. In method three, the bits for representing the filter parameters are given high priority and bits for representing the excitation signals are given low priority. Method four is an embedded coding technique based on two-stage CELP. The four methods were tested in combination with a simulated voice activity and queuing model and their performance is discussed. >

Study of voice packet reconstruction methods applied to CELP speech coding

The trend in today’s private digital networks is to integrate data with voice traffic. This offers considerable cost savings and convenience by allowing data and digitized voice to share the same transmission media, and much of the hardware and software. Code excited linear prediction (CELP) [1] is a speech coding method that has tremendous potential for providing the high quality that network voice applications demand for rates at and below 16 Kb/s. One difficulty with CELP is that it requires tremendous computational resources and memory for its codebook search. Just as there is a cost associated with the maximum bandwidth that a given network link can accommodate, there is a cost associated with the hardware necessary to support speech compression. With new telecommunications technologies promising to significantly reduce the cost of network bandwidth, the cost of the speech processing hardware may someday become the single most important consideration. Since digital signal processing chips (DSPs) and the fast memory chips supporting them consume a relatively major portion of a speech coding board’s real estate and power consumption, coding schemes that reduce the number of required DSPs and memory will significantly reduce the overall cost of a network.

Ronald G. Danisewicz

Papers

A Real-Time Full Duplex 16/8 kbps CVSELP Coder with Integral Echo Canceller Implemented on a Single DSP56001