Audio Training Files — Free WAVs

Audio training files are the calibrated reference signals behind honest listening tests: alignment tones, pink noise, sine sweeps, polarity and dynamic-range demos. Preview them in the browser or download them as uncompressed WAVs, and note that every file is synthesized locally, so nothing is uploaded.

How to use it

1. Filter the library by category. The tabs are All, Tones, Noise, Sweeps, Phase, Dynamics, Digital and Temporal, covering 12 files total.
2. Preview before you download. Preview Audio renders the selected signal through the Web Audio engine at your device’s sample rate and plays it through once.
3. Download the WAV you need. The download button writes a RIFF/WAVE file with 16-bit signed PCM samples at 44.1 kHz, exactly as the on-screen spec line states for each item.
4. Import into a DAW and check the level. Place the file on a track, confirm the peak or RMS reading on the DAW meter, and only then use it as a reference. A -20 dBFS peak sine reads -20 dBFS on a sample peak meter.
5. Set one monitoring level and leave it. Play the pink noise reference and trim your monitor controller so the noise sits where you want it. Mark that position; every later comparison inherits the same reference.
6. Work through the families in order. Tones for alignment, noise for spectrum and room work, sweeps for resonances, phase for wiring, dynamics and digital for resolution, and the Haas file for spatial widening.

The library at a glance

File	Signal	Duration	Primary use
1 kHz sine, -20 dBFS	Pure tone, peak calibrated	5 s	Alignment level, 0 VU reference
1 kHz sine, 0 dBFS	Full-scale sine at 0.999 amplitude	3 s	Check clipping and converter headroom
100 Hz sine, -20 dBFS	Low-frequency tone	5 s	Subwoofer and crossover checks
10 kHz sine, -20 dBFS	High-frequency tone	5 s	Tweeter output and top-end roll-off
Pink noise, -20 dBFS RMS	Equal energy per octave	10 s	Room EQ, RTA work, monitor calibration
White noise	Equal energy per hertz	5 s	Spectrum sweeps and filter behavior
Brown noise	6 dB per octave roll-off	5 s	Sub and bass-trapping evaluation
20 Hz to 20 kHz sweep	Logarithmic chirp	10 s	Modal resonances, buzzes, rattles
Polarity and phase check	Stereo pulse train, R inverted	5 s	Speaker wiring and phantom center
Dynamic vs compressed	DR14 against a limited half	6 s	Crest factor and transient training
8-bit truncation demo	Low-level tone quantized to 8 bits	5 s	Quantization grit without dither
15 ms Haas burst	Stereo click, right delayed	5 s	Spatial widening and pre-delay

Building a personal reference library

A reference library is only useful if it stays consistent. Keep the downloaded signals in one folder tree (tones, noise, sweeps, spatial, resolution), add a small set of reference tracks you know intimately, and treat the levels as part of the data.

• Choose three to five reference tracks and stop. Tracks you have heard hundreds of times outperform unfamiliar “audiophile” material because you carry an internal memory of how they should sound.
• Keep the site files as the calibration layer. The generated tones and noise never change and never carry a performance, which is what makes them comparable across sessions.
• Note the playback chain with each entry. Headphones or monitors, interface, any DSP, and the monitoring level. A reference library without chain notes turns every comparison into a guess.
• Use integrated loudness for program material. Peak dBFS tells you about headroom, and RMS tells you about average energy in a window, but loudness-normalized program comparison relies on gated LUFS as defined in ITU-R BS.1770. The on-site Loudness Meter reports that measurement.

Levels, formats and sample rates

The -20 dBFS alignment convention exists so that a reference tone hits a known mark while leaving 20 dB of headroom to full scale. On a digital meter, -20 dBFS peak with a full-scale ceiling at 0 dBFS means the reference sits 20 dB below clipping; broadcast practice in North America uses that number, while much of European broadcast aligns at -18 dBFS. Both work; the discipline is that everyone in a session uses the same one.

For level consistency, calibrate with pink noise rather than a sine. Noise exercises every band at once, so a single monitor trim covers the whole spectrum. One measured point: the pink noise file here renders at about -20.2 dBFS RMS, which matches its nominal calibration, while an offline check of the same generator math put the white noise file near -29.7 dBFS RMS and the brown noise file near -33.9 dBFS RMS. If you need absolute calibration, meter each noise file instead of trusting a nominal label, and remember that pink, white and brown noise are not equally loud at the same RMS.

Property	Value	Why it matters
Container	RIFF/WAVE, fmt + data chunks	Universally readable, no decoder quirks
Sample format	16-bit signed PCM, clipped at full scale	96 dB of theoretical dynamic range
Sample rate, downloads	44.1 kHz	22.05 kHz Nyquist, covers the audio band
Sample rate, preview	Your device’s native rate	Preview and download can differ at 48 kHz
Channel layouts	Mono, or stereo for phase and Haas files	Stereo files carry the inter-channel relationship
Duration	3 s to 10 s per file	Short enough to loop, long enough to measure

Compressed formats are the wrong tool for this job. MP3 and AAC encoders apply psychoacoustic masking and alter the waveform, so a codec can change the very phase or level relationship a reference file exists to demonstrate. Keep the WAV original and convert only for convenience playback, never for measurement.

Accuracy, limits and troubleshooting

• The preview is rendered at your device rate. A 44.1 kHz download played back after a 48 kHz preview can differ slightly in resampling behavior; the download itself is always the spec line stated on the card.
• Media players normalize. ReplayGain, Sound Check and “volume leveling” change the level of downloaded WAVs, which defeats level calibration. Use a DAW or disable normalization.
• OS processing defeats the test. Audio enhancements, AGC, spatial audio and loudness equalization alter level and phase. Turn them off for measurement sessions.
• Full-scale tones are a system stress test. The 0 dBFS 1 kHz file is designed to reveal clipping, not to be played at high volume for long. Keep sessions brief and comfortable, and never use test tones as a demo.
• A downloaded file cannot tell you what your room is doing. Room modes depend on position and construction; the Room Acoustic Test and Speaker Test cover that ground interactively.
• These are educational and measurement signals. They do not diagnose hearing loss. If you are concerned about your hearing, see an audiologist.

Related tools

• Noise Generator — generate pink, white and brown noise live at any duration.
• Sweep Generator — build custom frequency sweeps without leaving the browser.
• Loudness Meter — check integrated LUFS and true peak against ITU-R BS.1770.
• Colors of Noise Explained — understand 1/f, flat and 1/f squared spectra before you use them.