A Test of OggVorbis Encoder with a Simple Sine Sweep Signal

Nyaochi has found that a simple sine sweep signal at relatively low revel (-50dB) can reveals audible
degradation of .ogg files when prepared by OggVorbis RC2.

In this experiment, the Nyaochi's method has been applied to OggVorbis RC3, and the resultant files are
analyzed in some details along with MP3 and WMA8 codecs.

Used encoders are OggEnc v0.8 (libvorbis rc2) and v0.9 (libvorbis rc3) which are both released officially.

A sine sweep signal (160Hz`22kHz) of -51dB at peak for 60 sec are generated as an original .wav file
at log sweep mode by using CoolEdit Pro v1.2. The peak level of the original is enough to prevent the
measurements from any clipping.
The noise floor is at about -115dB, indicating that the original has enough signal_to_noise margin of 64 dB
for critical hearing tests.

[1] Just Listen

original(.wav, 44.1kHz, stereo): sweep.zip (2,896kB)

OggVorbis RC2: sweep_vorbis_rc2.ogg (153kB, encoded at -b 64, 44.1kHz)

OggVorbis RC2: sweep_vorbis_rc3.ogg (378kB, -b 64, 44.1kHz)

WMA8: sweep.wma (476kB, 64kbps, stereo)

MP3: sweep_lame.mp3 (470kB, -b 64 -m j --resample 44.1 --lowpass 14)

[2]Experimental

The above compressed files are decoded and analyzed by WaveLab v3.0.
Fig.1 shows wave form data taken at around 6.4kHz at which additional timbre and/or a sort of dropout can be
easily audible in OggVorbis RC2 and RC3.
The amplitude of .ogg files flutters, while .mp3 and .wma are constant.


Fig.1 Comparison of wave form at around 6.4kHz encoded by OggVorbis RC2, RC3, MP3(Lame) and WMA8 at 64kbps

Fig.2 is an enlarged view of 'fluttering'. It can be seen that, although sine wave itself is not distorted, the amplitude are
heavily modulated with random frequency.
Anyway, OggVorbis can not encode a simple sine sweep properly at this low bit rate range.


Fig.2 Enlarged view of wave form encoded with OggVorbis RC3.