Related;

This site;


Test tones

General

Each signal starts and ends with 5 ms silence and 10 ms or 10 period sinusoidal fade in- and out, which ever is longer. This is done to avoid clicking sounds at the beginning and end of the sound file;

 Fade in and out

The specified duration is between fade- in and out.

The signal starts with a positive going zero crossing, except when the sample-rate divided by the frequency is exactly two. In which case there is a 90° start phase offset. This avoids generating just zeros.
Because of the fade- in and out, the signal still starts and ends with zero though.

48000 samples per second

Peek level is 12 dB below clip level (25 %). This is 15 dB RMS below clip level (-9 dBm). Except for the 425, and 450 Hz files which are 3 dB lower. This effectively makes the files 14-bit.
400 Hz is an alternative to 1 kHz and was often used with TV test charts.
450 Hz is an old dial tone frequency. 425 Hz is current.
440 Hz is in fact a tuning fork frequency.
Each file is 60 seconds, except for the sweep which is 20 seconds. And the FLAC and OGG versions of the 440 Hz file are two minutes.

Wav

Little-endian 16-bit signed integer 48000 samples per second WAV files.

Flac

FLAC versions derived from the WAV files.

Ogg

OGG versions derived from the WAV files.

Other

Files with a higher signal level, sample-rate and resolution can be found here: High resolution test tones.

8000 samples per second

When generating signals with 8000 samples per second, there is also a phase offset whenever the sample-rate to frequency ratio is exactly a multiple of four. This offset is frequency / sample-rate * 180°.
The example below shows a signal with a frequency of 1/4 of the sample-rate;

Phase shifts 1:4
45° phase shift Without
phase
offset		 With
phase
offset
PhaseRel
Value		 PhaseRel
Value
0° 0 45° 0.7
90° 1135° 0.7
180° 0225°-0.7
270°-1315°-0.7
0° 0 45° 0.7

And a frequency of 1/8 of the sample-rate;

Phase shifts 1:8
22.5° phase shift Without
phase
offset		 With
phase
offset
PhaseRel
Value		 PhaseRel
Value
0° 0 22.5° 0.4
45° 0.7 67.5° 0.9
90° 1 112.5° 0.9
135° 0.7157.5° 0.4
180° 0 202.5°-0.4
225°-0.7247.5°-0.9
270°-1 292.5°-0.9
315°-0.7337.5°-0.4
0° 0 22.5° 0.4

This puts the samples on equal distances from zero, instead of at zero. Thus avoiding recurring zeros, which can't be properly decoded by alaw audio compression. It doesn't do zero; It just goes from 1 to -1!

Warning

Warning: These files are very loud!
Peek level is 3.05 dB below clip level (-6.06 RMS)!

A level of -3.01 dB is 0.5 * √2 (71 %). But with -3.05 dB these files can also be used to generate valid µlaw test files. Although µlaw is meant to compress 14-bit data, it's range is -8158 to 8158, not -8191 to 8191. That's why the signal is reduced by 8158 / 8191. And to compensate for rounding division is done by 8192 instead of 8191; 

 20 * log(0.5 * √2 * 8158 / 8192) ≈ -3.05 dB

This yields the following values;

Digital milliwatt
φ16-bit8-bitAlawµlaw
22.5° 8830 970xb40x9e
67.5° 21317 1160xa10x8b
112.5° 21317 1160xa10x8b
157.5° 8830 970xb40x9e
202.5° -8830 -970x340x1e
247.5°-21317-1160x210x0b
292.5°-21317-1160x210x0b
337.5° -8830 -970x340x1e

This signal is the equivalent of 0 dBm.
Note: With these 16-bit values the 8-bit values of both alaw and µlaw are identical.
Note: Dividing by 8191 instead of 8192 yields the same 8-bit values. In fact, slightly different 16-bit values will yield the same 8-bit results.
It may very well be that there is an official 16-bit version of the digital milliwatt. I have never been able to find one though.

Wav

Little-endian 16-bit signed integer 8000 samples per second WAV files.

A-law and µ-law

Alaw and µlaw versions derived from the above WAV files.

Alaw Wav

Alaw Raw

These have mime-type audio/g711-alaw.
Your browser may refuse to believe this though.

µlaw Wav

µlaw Raw

These have mime-type audio/g711-mlaw.
Your browser may refuse to believe this though.