HawkVoice™ codecs
Here are actual voice samples that were compressed and decompressed with the codecs included
with HawkVoice. The original u-law encoded samples were encoded and decoded back to u-law
for comparison.
This sample is about 5 seconds long, with a male and female speaker. Each sample is 43 KB and
available in .au or .wav format.
[.au] [.wav] Original u-law encoded
[.au] [.wav] 32 Kbps Intel/DVI ADPCM codec
[.au] [.wav] 13.2 Kbps GSM codec
[.au] [.wav] 4.8 Kbps LPC codec
[.au] [.wav] 4.5 Kbps CELP codec
[.au] [.wav] 3.0 Kbps CELP codec
[.au] [.wav] 2.4 Kbps LPC10 codec
[.au] [.wav] 356 bps - 2.4 Kbps VBR-LPC10 codec, averaged 1905 bps
[.au] [.wav] 2.3 Kbps CELP codec
[.au] [.wav] 1.8 Kbps OpenLPC codec
[.au] [.wav] 1.4 Kbps OpenLPC codec
[.au] [.wav] 2.4 Kbps fixed point LPC10 codec $
[.au] [.wav] 1.8 Kbps fixed point OpenLPC codec $ or LGPL
[.au] [.wav] 1.4 Kbps fixed point OpenLPC codec $ or LGPL
$ fixed point code is available for license, contact Phil Frisbie, Jr. for details
This sample is about 25 seconds long, with several male and several female speakers. Each sample
is 200 KB and available in .au or .wav format.
[.au] [.wav] Original u-law encoded
[.au] [.wav] 32 Kbps Intel/DVI ADPCM codec
[.au] [.wav] 13.2 Kbps GSM codec
[.au] [.wav] 4.8 Kbps LPC codec
[.au] [.wav] 4.5 Kbps CELP codec
[.au] [.wav] 3.0 Kbps CELP codec
[.au] [.wav] 2.4 Kbps LPC10 codec
[.au] [.wav] 356 bps - 2.4 Kbps VBR-LPC10 codec, averaged 1224 bps
[.au] [.wav] 2.3 Kbps CELP codec
[.au] [.wav] 1.8 Kbps OpenLPC codec
[.au] [.wav] 1.4 Kbps OpenLPC codec
[.au] [.wav] 2.4 Kbps fixed point LPC10 codec $
[.au] [.wav] 1.8 Kbps fixed point OpenLPC codec $ or LGPL
[.au] [.wav] 1.4 Kbps fixed point OpenLPC codec $ or LGPL
$ fixed point code is available for license, contact Phil Frisbie, Jr. for details
Here are some performance results calculated with codectest.exe, version HVDI 0.9, on a
Pentium III system using the short five second voice sample above:
CPU cycles per second for 8KHz sample rate sound.
Compression % is compared to 16 bit PCM.
encode decode compression
u-law: 42K 40K 50%
ADPMC: 407K 330K 75%
GSM: 2.0M 950K 89.7%
LPC: 2.5M 1.0M 96.3%
CELP 4.5K: 24-52M* 4.4M 96.5%
CELP 3.0K: 25-47M* 4.0M 97.7%
LPC-10: 6.4M 3.5M 98.1%
CELP 2.3K: 24-45M* 3.8M 98.2%
OpenLPC 1.8K: 2.9M 1.8M 98.6%
OpenLPC 1.4K: 2.9M 1.9M 98.9%
*Note on CELP encoding: CELP uses a codebook
of 256 speech patterns. The CELP encoding
performance listed shows figures from a codebook
search of 32 up to the full 256 entries.
Here are some performance results for fixed point codecs on a Pocket PC with a PXA250 400MHz
X-Scale (ARM) CPU using the long 25 second voice sample above:
CPU cycles per second for 8KHz sample rate sound.
encode decode
LPC-10: 19.0M 13.2M
OpenLPC 1.8K: 8.1M 9.8M
OpenLPC 1.4K: 8.2M 9.8M
HawkNL, HawkNLU, and HawkVoice are trademarks of Phil Frisbie
to help protect these open source projects.