Make a WAV File Quieter Without Uploading It

How it works

How it is done

1. Check that the file is a little-endian RIFF or RF64 WAVE container, then walk its padded chunk list to find exactly one fmt block and one data block. Read RF64's 64-bit data length from ds64 when required.
2. Read the channel count, sample rate, block alignment, bit depth and format tag. Unwrap WAVE_FORMAT_EXTENSIBLE only when its valid precision fills the sample container; refuse compressed codecs, big-endian RIFX and inconsistent frame widths.
3. Decode each 8-, 16-, 24- or 32-bit PCM sample, or each 32- or 64-bit IEEE-float sample, and measure the largest positive and negative amplitude. Reject non-finite float values rather than propagating NaN or infinity.
4. Convert the requested decibel change to an amplitude multiplier with multiplier = 10 raised to gain dB divided by 20. When clipping prevention is on for a boost, reduce that multiplier to the smaller signed headroom limit.
5. Copy the complete source file in bounded pieces. In the copy's data chunk, multiply each channel sample by the same factor, round integer PCM symmetrically to the nearest code, and clamp any over-range result at digital full scale.
6. Report the actual gain, measured peaks, number of samples that required limiting and unchanged duration. Offer the copied container as a separate WAV download; the source file remains untouched.

What it assumes

• Supported audio is little-endian integer PCM at 8, 16, 24 or 32 bits, or IEEE float at 32 or 64 bits. Compressed WAV codecs and big-endian RIFX require a decoder or byte-order conversion and are refused.
• WAVE_FORMAT_EXTENSIBLE is supported when its valid-bit count is zero or equals the container width. Reduced-precision samples such as 24 valid bits left-aligned in 32-bit containers are refused rather than interpreted by guesswork.
• The same constant factor is applied to every sample in every channel, preserving channel balance and relative dynamics. This is peak gain adjustment, not LUFS loudness normalisation, compression, limiting over time, resampling or dithering.
• Peak prevention caps only a requested positive gain. If an IEEE-float source is already outside the nominal -1 to +1 range, the applied boost becomes 0 dB and those existing over-range samples are still clamped and counted; the tool does not silently turn a boost into attenuation.
• Integer PCM is rounded to its nearest representable sample code after multiplication. At half-way cases, equal positive and negative magnitudes round away from zero so the operation does not introduce sign-dependent rounding.
• RIFF chunks and their positions remain byte-for-byte unchanged outside the data payload. That keeps cues and channel metadata in place, but any embedded peak, loudness, checksum or analysis metadata can describe the old samples and is not recalculated.
• Input is capped at 100 MB. The browser holds the source and copied output at once, while the scan, copy and rewrite loops yield in bounded pieces so progress and Cancel remain responsive.