Can a lossless file be larger than the input?

Yes, especially when overhead exceeds the available savings.

Does converting a lossy recording to lossless restore missing detail?

No. It preserves the data supplied to that conversion.

Must two files have identical hashes to contain identical decoded audio?

No. Container and metadata differences can change file bytes while audio samples remain the same.

AI Glossary / MLOps and deployment / Advanced

Lossless compression

Definition

Lossless compression represents data in a form from which the defined original data can be recovered exactly. Its goal is usually to reduce storage or transmission requirements by exploiting structure or statistical regularity. Exactness applies to the data being encoded: for a general file compressor this may mean every file byte, while a media codec may define its promise around audio samples or image values rather than every surrounding metadata byte.

What it means

Lossless compression saves space without deliberately discarding information from its defined input. It is different from a lossy process that accepts changes in exchange for smaller representations. However, the label does not guarantee that every input becomes smaller, that all metadata is retained or that the original recording was accurate. A lossless copy of a flawed measurement preserves that flaw. It also cannot restore details removed by an earlier lossy conversion simply because the result is later saved in a lossless format.

How it works

Methods can describe repeated strings, encode common symbols efficiently or predict values and store exact corrections. The decoder reverses the specified representation to recover the input. Compression effectiveness depends on the data and the overhead needed to describe it. Some inputs expand. This is unavoidable for a general exact scheme: there are fewer shorter binary strings than all strings of a given length, so distinct inputs cannot all be assigned distinct shorter representations. A practical format also defines framing and enough information to decode its contents unambiguously.

Three practical examples

A project team compresses source files for transfer and compares the recovered bytes with the originals.
An audio archive uses a lossless codec for supported PCM samples while separately verifying which metadata must be retained.
A data engineer compresses a repetitive dataset and measures total stored size, encoding time and restoration time before choosing settings.

Think of it this way

Imagine replacing repeated instructions in a document with a shared abbreviation list. Readers can restore every original instruction if the list and the replacement rules are complete. The list itself takes space, so abbreviating a tiny document may make the package larger. Lossless compression follows the same broad accounting principle: the recovery instructions count too. Unlike an informal summary, the compressed representation must preserve enough information to reconstruct the defined input exactly, rather than merely communicate its overall meaning.

Worked example

Consider an original toy run-length format for uppercase ASCII letters. Each run is stored as one count byte and one letter byte, with run lengths limited to 1 through 255. The twelve-byte input “AAAAAAAABBBB” has runs of eight A characters and four B characters. Its two count-letter records occupy four bytes, excluding any outer header. That saves eight bytes, or about 66.7 percent of the original size. The four-byte input “ABAB” instead requires four records and occupies eight bytes, doubling its size. Both examples decode exactly. The contrast illustrates why reversibility does not imply a reduction for every input; the toy format is not a claimed standard archive format.

Illustrative case study

A fictional research group migrates audio recordings and checks only the total file hashes after conversion. The hashes differ, so the group initially concludes that the lossless codec changed the sound. It then separates the tests: decoded sample values are compared directly, and required metadata is inventoried independently. The team finds identical samples but different container details in its example workflow. It updates the migration requirements to specify both sample preservation and metadata retention. This scenario illustrates how a vague promise to keep “the original” can hide multiple distinct preservation requirements.

Common misconceptions and mistakes

A lossless codec does not guarantee that the compressed file is smaller than every alternative or even smaller than the input. Higher compression settings need not mean higher reconstructed quality when both outputs recover the same input exactly. Another mistake is equating identical decoded content with identical container bytes. Conversely, a file that looks or sounds unchanged is not necessarily bit-exact. Compression is also distinct from encryption, backup and error correction. A compressed file can still be readable by others, lost with its storage device or damaged beyond recovery.

Business use

Lossless compression is useful where exact records, reproducible datasets or original media values matter. Define the preservation target before selecting a tool: entire files, decoded samples, pixel values, metadata or some combination. Measure complete output size and restoration performance on representative material. Verify recovered data against originals and include decoding support in retention planning. Keep independent backups where required; a smaller copy is not inherently a safer copy. The practical benefit comes from reducing resource use while preserving clearly specified information and a reliable recovery path.

Check your understanding

How many bytes did the two-run toy example occupy before outer framing?

Answer: Four.

Why did “ABAB” expand?

Answer: Each one-byte letter required a two-byte run record.

What must an audio migration test specify?

Answer: Whether exactness concerns decoded samples, complete file bytes, metadata or several of these together.

Related terms

Lossy compression
Entropy coding
Data integrity
dictionary compression
predictive coding
container format

Learn next

Compress several differently structured inputs and record actual output sizes. Restore them and compare the defined data directly. For a media example, check decoded content separately from metadata so a container change is neither mistaken for signal loss nor allowed to hide missing information.

Sources

Xiph.Org Foundation, FLAC FAQ: https://www.xiph.org/flac/faq.html The run-length calculations, abbreviation analogy and migration case are original educational illustrations. The FAQ supports the distinction between audio-sample preservation and surrounding file details; the toy format is independent of FLAC's encoding method.

Previous: Loss surface Next: Lossless join decomposition