Bit-perfect playback
Bit-perfect playback preserves a defined decoded sample sequence through a specified part of the audio output path.
“Bit-perfect” is meaningful when it names a reference sample sequence and an output boundary. For a lossless music file, a useful claim is that the decoded PCM sample values reach a particular digital output unchanged, with the same channel order, sample rate and intended sample count.
The compressed file bytes are not the reference PCM. A FLAC file can be decoded to WAV with different file bytes while preserving every audio sample. Comparing the hashes of those two files would answer the wrong question.
Identical values can have different packing
Digital interfaces need not use the same sample container as the source. A 16-bit value can be represented exactly in a 24-bit signal domain by scaling its integer representation by 256, leaving eight low bits zero.
| Signed 16-bit integer | Corresponding signed 24-bit integer | Normalized amplitude |
|---|---|---|
| −32,768 | −8,388,608 | −1 |
| −1 | −256 | −1/32,768 |
| 0 | 0 | 0 |
| 1 | 256 | 1/32,768 |
| 32,767 | 8,388,352 | 32,767/32,768 |
Dividing the second column by 256 recovers the first exactly. The raw byte streams differ, but the represented amplitudes agree. A comparison therefore has to account for valid-bit alignment, container size, byte order and interleaving.
Similarly, converting ordinary signed 24-bit PCM into normalized 32-bit floating point can represent those values exactly. The presence of a floating-point buffer does not by itself prove a loss of precision. Subsequent processing and conversion determine what reaches the output.
If a test requires literal byte identity in one specific PCM format, it should say so. That stricter claim differs from preserving the same sample values through reversible repacking.
Operations that change the sequence
A gain other than unity changes sample amplitudes. ReplayGain, an equalizer, a downmix, spatial rendering or a limiter also changes the sequence. Resampling computes values on a different time grid. Dither intentionally adds noise during a precision reduction.
Those processes can serve a useful playback purpose. A result can sound appropriate without satisfying the preservation claim. Conversely, an unchanged sequence can still be delivered late, to the wrong speaker, or into a reproduction system with a problem.
For a lossy file, the decoder's PCM output can be the stated reference for an output-path test. That does not mean the codec has recovered the exact pre-encoding samples. The claim begins after that particular decode.
Exclusive access narrows the route
Exclusive-mode output can bypass an operating system's shared mixing stage. It makes one class of sample modification easier to avoid, but it does not disable every player filter, driver process or device DSP.
Shared access can also preserve a particular stream under suitable conditions: matching formats, unity gain, no additional contributing audio and no modifying processing. Whether those conditions hold must be established on the actual route. The mode name alone is insufficient evidence in either direction.
An Emby server's Direct Play status concerns media delivery. A client can receive the original FLAC file, decode it accurately, and then apply volume or resampling. Server-side retention and client-side sample preservation are different observations.
A useful verification boundary
A digital capture at the output boundary can be decoded into a common comparison format, aligned in time, and compared against the reference PCM. Alignment should account for a fixed delay without silently deleting different audio. The check should preserve sample rate, channel identity and count, and report any nonzero differences.
An analog recording after a DAC adds the behavior of converters, clocks and noise; it cannot establish literal digital sample identity. A device that receives special encoded data may offer its own recognition test, but that is a test of that supported route rather than a general listening judgment.
State the boundary with the result: for example, “decoded 16-bit, 44.1 kHz stereo values preserved to the captured digital output after reversible 24-bit packing.” That description tells a reader exactly what was established.