Cut a WAV Without Losing Quality

How it works

How it is done

1. Check that the first twelve bytes identify a little-endian RIFF or RF64 WAVE file. A big-endian RIFX file is refused because copying its samples into a little-endian container would change their meaning.
2. Walk the RIFF chunk list, respecting the pad byte after an odd-length payload, until the fmt and data chunks have both been found. For RF64, read the 64-bit audio length from its ds64 chunk.
3. Read the channel count, sample rate, block alignment and bit depth from fmt. If the format is WAVE_FORMAT_EXTENSIBLE, unwrap its sub-format GUID to distinguish PCM from IEEE float.
4. Verify that the format is fixed-width PCM or IEEE float and that the declared block alignment equals channels multiplied by bytes per sample. A compressed or internally inconsistent file is refused rather than cut at a guessed boundary.
5. Divide the data length by the block alignment to get the exact number of complete sample frames and, from that, the recording length.
6. Multiply each requested time by the sample rate and round to the nearest whole frame. Clamp an end beyond the recording to its final frame; refuse a start at or beyond the end.
7. Copy the selected frame range in 4 MB pieces into a fresh WAV. The format block is preserved, a new fact count is written for float audio, and the RIFF and data sizes are recalculated.
8. Leave metadata chunks out of the result because their cue positions, loop points and time references would describe the untrimmed recording. Offer the new metadata-free WAV as a download.

What it assumes

• PCM and IEEE-float WAV files are supported, including WAVE_FORMAT_EXTENSIBLE and RF64. Compressed WAV encodings and big-endian RIFX are not, because cutting either safely requires decoding or byte-order conversion.
• A requested time can fall between sample frames. It is rounded to the nearest complete frame, so the largest timing adjustment is half a frame — about 0.011 milliseconds at 44.1 kHz.
• An end time beyond the file is clamped to the last complete frame. A start at or beyond that frame leaves no playable audio and is refused rather than producing an empty file.
• The sample rate, channel count, bit depth, format block and retained sample bytes are copied. There is no gain change, fade, resampling, dither or codec pass.
• Metadata is deliberately not copied. Broadcast time references, cue points, regions, loops, labels, artwork and comments would otherwise point into the old timeline or disclose details the visitor did not intend to keep.
• A data chunk must contain a whole number of sample frames and fit inside the file. A partial last frame or a chunk length past the end is treated as truncation rather than silently repaired.
• The whole input and the trimmed output exist in memory at once. Input is therefore capped at 256 MB, and a bounded RF64 result is written as an ordinary RIFF WAV.