Efficient speech and channel effects using a modified low bit rate speech codec

(Or, the architecture behind TinySpeech for Games™)


Background

Warning! Technical information follows. You may wish to skip this section and go right to The Problem
.
TS is a low complexity, variable bit rate, variable quality speech codec, or speech compressor. It uses algorithms that are patent free, or patent expired. It currently supports 8 KHz sample rate only, but 16 KHz is a planned addition for higher quality speech. It is a CELP type codec, but it does not use a fixed code book. The encoder block diagram is basically this:

Speech in -> pitch determination -> gain determination -> LPC analysis -> LPC to LSF conversion -> residual determination -> create codebook and gain->add control bits

Speech in: 16 bit PCM samples, 8 Khz sample rate.
Pitch determination: Pitch is quantized to a 6 bit value.
Gain determination: Gain is quantized to a 6 bit value.
LPC analysis: A 10 pole Linear Predictive Coding filter is created.
LPC to LSF conversion: The LPC parameters are converted to Line Spectral Frequencies and quantized to 24 bits.
Residual determination: The LSF is converted back to LPC, the speech is inverse filtered with the LPC to create the residual speech (the decoding error).
Create codebook and gain: the codebook is created (variable number of bits), and two gain values (first and second half frame) are quantized at 4 bits each.
Add control bits: various other control bits are added to the encoded frame to control comfort noise, voiced/unvoiced flags, etc.

TS is always variable bit rate, and it encodes silence frames with comfort noise. Unvoiced frames are encoded with fewer LSF bits than voiced frames, and the code book can be varied in the number of bits per frame or omitted completely. The maximum bit rate can be preset, and the code book can also be disabled. This architecture was designed from the start to support TSGs.

Next: The Problem


Prologue and Introduction
Background
The Problem
The Solution
The Solution continued and Conclusion


TinySpeech and TinySpeech for Games are trademarks of Phil Frisbie, Jr.
Page last modified: 27 February 2006
© 1998-2006 Hawk Software