Dither

In-Depth Explanation

Dither is a low-level noise signal added to an audio recording before reducing its bit depth (for example, converting 24-bit to 16-bit). Without dither, the bit-depth reduction produces quantization error, which manifests as harsh, correlated distortion on quiet passages and fade-outs. Dither replaces this distortion with a smooth, uncorrelated hiss that the human ear perceives as a natural noise floor rather than a sonic defect.

How Dither Works

Digital audio stores amplitude as integer values. A 24-bit recording has 16,777,216 possible amplitude values per sample. A 16-bit recording has 65,536. When you reduce 24-bit audio to 16-bit, the converter must round each sample value to the nearest available 16-bit value. This rounding process is called quantization.

The error between the original value and the rounded value is called quantization error. When the audio signal is loud, this error is random and inaudible. When the signal is quiet (during a fade-out, a reverb tail, or a silent passage), the error becomes correlated. The signal repeatedly rounds to the same incorrect value, producing harmonic distortion that follows the signal. This distortion is harsh, buzzy, and far more noticeable than a smooth noise floor.

What Dither Does

Dither adds a very low-level random noise signal (typically at the noise floor level of the target bit depth) to the audio before quantization. This noise forces the quantization error to become random instead of correlated. Instead of harsh distortion, you get a gentle, analog-style hiss.

The tradeoff is simple: you add a tiny amount of noise to eliminate a more objectionable form of distortion. The noise is typically around -93 dBFS for 16-bit dither, which is well below the noise floor of any playback system and inaudible in normal listening conditions.

Types of Dither

Most DAWs offer several dither algorithms. The differences are in the noise spectrum and whether noise shaping is applied.

1. TPDF (Triangular Probability Density Function): The standard dither algorithm. Adds white noise with a triangular distribution. This is the mathematically correct minimum dither that fully decorrelates quantization error. Used by Pro Tools, Logic Pro, and most DAWs as the default.

2. Rectangular: Adds less noise than TPDF but does not fully eliminate correlated distortion. Rarely recommended.

3. Noise-shaped dither: Applies an EQ curve to the dither noise, pushing it into frequency ranges where human hearing is less sensitive (typically above 15 kHz). This allows the dither to be louder in those ranges without being perceived, while keeping the audible mid-range quieter. Algorithms like POW-r (Psychoacoustically Optimized Wordlength Reduction) and iZotope's MBIT+ offer multiple noise-shaping curves optimized for different material. Noise-shaped dither can preserve more detail in the audible range than flat TPDF.

When to Apply Dither

Dither should be applied exactly once, at the very last step of production, when reducing bit depth. The most common scenario is exporting a 24-bit master to a 16-bit WAV for CD or a distributor that requires 16-bit files.

You should not apply dither when:

• Bouncing between 32-bit float and 24-bit within your DAW (32-bit float has sufficient resolution that quantization error is inaudible)
• Exporting at the same bit depth as your session
• Sending stems to a mixing or mastering engineer (keep them at 24-bit or 32-bit float)

If you apply dither and then process the audio again (like adding another plugin or changing the volume), the dither is altered and loses its effectiveness. Dither is always the final operation before the file is written.

Real-World Example

You mix a track at 24-bit in Ableton Live. The final mix includes a long reverb tail that fades from -30 dBFS down to -90 dBFS over 8 seconds. You export the master to 16-bit for CD delivery without applying dither.

On playback, the reverb tail sounds gritty and metallic as it fades below -48 dBFS. Below -60 dBFS, the tail breaks up into a buzzing, stepped sound instead of smoothly fading to silence. This is quantization distortion. The 16-bit quantizer is rounding the quiet reverb samples to the nearest available value, and the error pattern creates audible harmonics.

You re-export with TPDF dither enabled. The reverb tail now fades smoothly to silence. There is a very faint hiss audible in the final 2 seconds of the fade, but it sounds like natural analog noise, not digital distortion. The hiss is at approximately -93 dBFS, which is below the noise floor of any consumer playback system.

You try iZotope's MBIT+ dither with noise shaping instead. The fade is equally smooth, but the faint hiss is even less noticeable because the noise-shaping curve pushes the dither energy above 15 kHz, where human hearing is less sensitive.

Why It Matters for Independent Artists

Most independent artists deliver 24-bit WAV files to distributors like DistroKid and CD Baby, which accept 24-bit directly. In that case, dither is not needed for the delivery file. But dither becomes relevant in three common situations:

1. CD production: CDs require 16-bit audio. If your master is 24-bit, you must downconvert to 16-bit with dither. Without it, quiet passages and fade-outs will contain audible quantization distortion.

2. Reference mixes and demos: If you bounce a rough mix to 16-bit MP3 or 16-bit WAV to send to a collaborator, enable dither on the export. Even on compressed formats, the bit-depth reduction happens before encoding.

3. Mastering for streaming: Some older distribution pipelines still require 16-bit delivery. If your distributor specifies 16-bit, apply dither during the final export from your mastering session.

The rule is simple: any time you reduce bit depth, apply dither. Any time you do not reduce bit depth, do not apply dither. Never apply dither twice. For a walkthrough of the export process, read our mastering for streaming platforms guide and our mixing vs. mastering article.

Related Terms

• Bit Depth - The property being reduced when dither is applied
• Mastering - The final stage where dither is typically applied during export
• Noise Floor - The level at which dither noise sits in the final audio file
• Sample Rate - Often confused with bit depth, but sample rate reduction does not require dither
• DAW (Digital Audio Workstation) - The software where dither is applied during export