Browse Prior Art Database

IMBE PITCH TRACKING IMPROVEMENTS

IP.com Disclosure Number: IPCOM000008234D
Original Publication Date: 1997-Jun-01
Included in the Prior Art Database: 2002-May-29
Document File: 4 page(s) / 195K

Publishing Venue

Motorola

Related People

John Monson: AUTHOR [+2]

Abstract

IMBE (Improved Multiband Excitation) is a speech coding algorithm developed by DVSI (Digital Voice Systems Inc.). A version of this algo- rithm, operating at 4350 bits per second, is specified for use in APCO Project 25 compliant systems'. The pitch tracking algorithm in the encoder plays a significant role in synthesizing audio from the microphone samples and therefore is vital to overall IMBE audio quality. It is felt that audio quality in Motorola SR3.0 radio systems in real-world channel conditions can be improved by making minor changes to the version 1.5.0 IMBE pitch tracking algorithm.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 34% of the total text.

Page 1 of 4

MO-LA Technical Developments

@

IMBE PITCH TRACKING IMPROVEMENTS

by John Monson and Aaron Smith

INTRODUCTION

  IMBE (Improved Multiband Excitation) is a speech coding algorithm developed by DVSI (Digital Voice Systems Inc.). A version of this algo- rithm, operating at 4350 bits per second, is specified for use in APCO Project 25 compliant systems'. The pitch tracking algorithm in the encoder plays a significant role in synthesizing audio from the microphone samples and therefore is vital to overall IMBE audio quality. It is felt that audio quality in Motorola SR3.0 radio systems in real-world channel conditions can be improved by making minor changes to the version 1.5.0 IMBE pitch tracking algorithm.

BACKGROUND

  The following is a summary of the IMBE 1.5.0 pitch tracking algorithm for the fixed-point DSP56K. The algorithm is used to identify the pitch of the current frame with l/2 sample accuracy (62.5 psec). The algorithm attempts to find the best pitch track starting or terminating in the current frame. This is done using the following steps:

A. Look forward and backward 2 frames (40 ms).
B. Set the allowable frame to frame pitch deviation to 20%.

C. Find the minimum error from the current frame to 2 frames in the future.

D. Find the minimum error from 2 frames in the past to the current frame.

E. Compare the error between the two paths and choose the pitch estimate for the current frame.

  Initial pitch estimates are restricted to the range PO = (21,21.5, .., 121.5, 122), where P0 is based on 8 KHz audio samples.

  The look-ahead pitch tracking algorithm (step C) attempts to preserve pitch continuity between the current and future speech frames. It computes forward pitch estimates based on future (buffered) frames, and a forward cumulative error function, CE,(P,), from the forward pitch estimates. The forward pitch estimate, Pp, is defined as the value of P from the set of P, that minimizes CI$(P,,). In addition, P, is chosen only after integer sub-multiples of P, are considered.

  Look-back pitch tracking (step D) uses previous frame pitch values and calculates a backward cumu- lative error function, Cl&(P,). The backward pitch estimate, P,, is defined as the value of P from the set of P0 that minimizes ClQP,).

  Once PF, P,, CE,(P,,) and CE,(P,) have been computed, the initial pitch estimate, P,, is chosen according to the following rules:

IF CIQP,) < 0.48,

THENP,=P,

ELSE IF ChPo) <= CWJ',),

THENP,=P,

ELSE P, = P,

  where the initial pitch estimate, P,, is a member of the set {21, 21.5, .., 121.5, 1221. This relation- ship can be seen in the following diagram:

e M.atmda, Inc. I997 196 June 1997

[This page contains 15 pictures or other non-text objects]

Page 2 of 4

m MO-LA Technical Developments

CEFIPI

CE FIP)

'B

PF

/

\

CEB (PI

0.30 CEB(P)

Fig. 1 Current IMBE

IMPROVEMENTS

  It is possible to make subjective improvements to the IMBE 1.5.0 pitch tracking algorithm, with very minor code changes. This is accomplished by changing the threshold in...