AUD · Audio tools

Bit Depth Converter

Shareable link

Settings are written to the URL as you change them. Nothing differs from the defaults yet.

Every digital file has a word width, and going from a wider one to a narrower one throws away the bits that do not fit. What separates a good conversion from a bad one is what happens to the part that is thrown away.

The grid

At 16-bit there are 65,536 possible sample values. A sample that falls between two of them has to become one of them, and the difference is the quantisation error.

If the sample values are simply rounded, that error is a deterministic function of the input, which means it is a distortion of the signal rather than noise added to it. On a fade to silence, the distortion is what you hear last.

Dither

Adding a small amount of noise before rounding breaks the relationship between the signal and the error. The error becomes noise, which is far less objectionable than distortion at the same level.

Rectangular dither is one quantisation step wide and fixes the average error. Triangular dither is the sum of two rectangular sources, two steps wide, and fixes the variance as well. That last part is what stops the noise floor from moving with the music, and it is why triangular is the default here and in every mastering chain.

The dither level control exists for the rare case where you want less than the textbook amount. Below about -6 dB the dither stops fully decorrelating the error and you get a mix of noise and distortion.

Noise shaping

The quantisation error can be filtered before it is fed back, which moves the noise energy up in frequency. The total noise gets louder; the noise where the ear is sensitive gets quieter.

Light is a gentle second-order tilt that is safe at any sample rate. Heavy is the Lipshitz E-weighted curve, which buys the most perceived resolution and puts a lot of energy above 15 kHz.

Shaped noise only stays shaped until someone processes the file again. Use it on the last step before delivery, not in the middle of a chain.

Getting the depth into the file

Quantising the samples is only half the job. The encoder that writes the file also picks a word width, and if it picks 16-bit after the step has carefully placed everything on the 24-bit grid, the work is undone on the way out.

This tool quantises to exactly the grid the PCM encoder will use, and locks the export to the matching WAV codec, so the encoder’s own rounding changes nothing. Check the result in the metadata viewer: the header should report the width you chose.

8-bit

The grid is 256 steps and the noise floor is around -48 dBFS, so dither is not subtle here. It is a sound rather than a format choice, which is the honest reason to reach for it.

Frequently Asked Questions

A WAV whose samples sit on the 8, 16, or 24-bit integer grid and whose header declares that width. It reads WAV, MP3, M4A, AAC, Ogg, Opus, FLAC, AIFF, and WebM audio.

Rounding on its own produces an error that follows the signal, and a signal-correlated error is heard as distortion rather than noise. It is worst where it matters most: fades, reverb tails, and quiet passages. Adding a little noise before rounding decorrelates the error, trading distortion for a steady hiss.

Triangular. Rectangular dither decorrelates the error's mean but leaves its variance moving with the signal, which is audible as noise that breathes. Triangular costs about 4.8 dB more noise floor and removes the modulation completely.

Going to 16-bit for a final master, where it buys several apparent bits in the midrange by pushing noise up where the ear is least sensitive. Skip it if the file will be processed again afterward, because shaped noise stops being shaped once someone resamples or re-encodes it.

No, and it says so. A 32-bit integer grid is finer than the float samples being quantised, so the step passes the audio through untouched and the export simply writes 32-bit words.

Explore Our Tools

Browse all tools