Stream Copy vs Re-encoding When Splitting MP3 Audiobooks

“Lossless MP3 splitting” sounds contradictory because MP3 is a lossy format. In this context, it means something narrower and useful: divide the file without applying another lossy encode.

That is what stream copy is designed to do.

What happens during stream copy

A normal transcode has three broad stages:

compressed MP3
     ↓ decode
uncompressed audio
     ↓ encode
new compressed MP3

Stream copy avoids the middle conversion:

compressed MP3 packets
          ↓ copy selected ranges
new MP3 chapter files

FFmpeg describes -c copy as copying stream packets without decoding, filtering, or encoding. This makes it fast and prevents quality loss from a new encode, although it is not compatible with every source and output combination. See the official FFmpeg stream-copy documentation.

Why re-encoding can reduce quality

MP3 encoding removes information to reduce file size. Decoding that MP3 and encoding it again asks a lossy encoder to make a new set of decisions from material that has already been simplified.

The result may still sound perfectly acceptable for spoken-word listening, especially at a suitable bitrate. But increasing the output bitrate does not restore discarded detail. A 64 kbps source encoded again at 320 kbps becomes larger, not more authentic.

Why stream copy is not always the safest output

Stream copy depends on the source packets, timestamps, seek points, codec, and destination container cooperating.

It may be unsuitable when:

• chapter boundaries fall between convenient packet positions;
• timestamps are malformed or inconsistent;
• the source uses unusual variable-bitrate behaviour;
• the destination player is sensitive to the resulting files;
• audio must be filtered, resampled, or normalised;
• a codec or container change is required.

That is why a good splitter needs a fallback rather than promising that packet copying works for every audiobook.

How ChapterSplit handles this on macOS

The current Mac engine attempts stream copy first. For a compatible source, this gives you fast splitting without another audio encode.

If that attempt fails, ChapterSplit can fall back to re-encoding. The fallback can use a fixed bitrate or VBR quality strategy, depending on the selected configuration. Cancellation is kept distinct from failure so cancelling a job does not unexpectedly begin a fallback encode.

Cover art is explicitly mapped and marked as an attached image so it is not accidentally trimmed away from chapters that begin later in the source.

How ChapterSplit handles this on Windows

The current Windows engine creates MP3 chapter audio with the libmp3lame encoder. Its primary path uses high-quality VBR settings.

For a known class of VBR encoder failure, it retries using safer fixed-bitrate and sample-rate settings. The default configuration also supports preserving cover art, adding track numbers, adding title tags, and choosing fixed or relative silence thresholds.

This is not stream copy, so the clearest Windows marketing language is controlled high-quality MP3 output with compatibility fallback, not “lossless splitting.”

Which approach should you choose?

A quality-first workflow

1. Keep the original audiobook.
2. Test one representative file.
3. On Mac, start with the normal stream-copy-first workflow.
4. On Windows, use the standard high-quality mode unless compatibility requires the safer mode.
5. Inspect chapter beginnings and endings.
6. Compare total duration with the source.
7. Test the output on the device where you will actually listen.
8. Process the rest of the library only after the sample works.

What to listen for

Use headphones for the validation pass and check:

• clipped opening consonants after a split point;
• duplicated words between neighbouring files;
• missing audio at the end of a chapter;
• unexpected clicks or gaps;
• cover art disappearing from later tracks;
• files that refuse to seek correctly in the target player.

You do not need to listen to the entire book. A few seconds on each side of several boundaries provides a much better signal than comparing file sizes alone.

When file size grows unexpectedly

If a re-encoded chapter collection is much larger than the source, check the selected bitrate. Matching a source's approximate bitrate is usually more rational than automatically choosing the largest number available.

For spoken-word material, the source recording, codec, channel layout, and listening environment matter more than a marketing label such as “maximum quality.”

The wording to trust

Accurate descriptions are more useful than absolutes:

• Good: “Avoids another lossy encode when stream copy succeeds.”
• Good: “Falls back to encoding for compatibility.”
• Good: “Preserves the quality already present in the source stream.”
• Misleading: “Makes MP3 lossless.”
• Misleading: “Improves the original audio quality.”
• Misleading: “Always cuts at sample-perfect chapter boundaries.”

Related guides

• How to Split a Long MP3 Audiobook into Chapters on Mac or Windows
• Embedded Chapters vs Silence Detection: Which Mode Should You Use?
• Complete ChapterSplit guide
• How to Convert M4B to MP3 Without Losing Chapters