Audio files are not defined only by .mp3 or .wav. A recording can be described by its sampling rate, channel layout, bit depth, compression codec, bitrate, loudness, clipping headroom, metadata and intended use. A WAV can hold uncompressed PCM, while a FLAC can preserve the same samples losslessly at a smaller file size. An AAC or Opus export can sound excellent for its purpose but cannot be transformed back into an exact original that was never retained. Understanding these distinctions prevents wasted storage and disappointing conversions.
Best practice: Archive an appropriate uncompressed or lossless master when feasible, and create distribution copies suited to playback compatibility, bandwidth and user expectations.
1. What a sample represents
Digital audio measures a continuously varying sound signal at discrete time intervals. A sample rate of 48,000 samples per second (48 kHz) means each channel records 48,000 sample positions every second. For ordinary PCM, bit depth determines how finely each sample's amplitude is represented; 16-bit integer PCM and 24-bit integer PCM have different storage needs and ideal quantization-noise behavior.
This does not mean 48 kHz audio always contains 48 kHz audible frequencies. The Nyquist relationship means a properly band-limited signal sampled at 48 kHz can represent frequencies below half that sampling rate, approximately 24 kHz, with practical filtering margins. High-rate recording has valid uses in production pipelines, but a higher number alone does not guarantee audibly better results.
Changing a file header from 44.1 kHz to 96 kHz cannot recover frequencies the source never captured. Real resampling requires signal processing and appropriate filters; careless conversion may introduce artifacts.
2. PCM file-size arithmetic
For uncompressed interleaved PCM, a useful approximation is:
bytes = seconds × sample rate × channels × (bits per sample ÷ 8)
For a 60-second stereo 48 kHz 16-bit recording, that yields 11,520,000 bytes, approximately 11.52 MB in decimal units. WAV headers and optional metadata add some extra bytes; the calculation describes PCM payload, not an exact file size for every WAV variant. With 24-bit PCM, the same recording's sample payload would be 17.28 MB. This is why uncompressed masters grow quickly during long shoots or multitrack sessions.
Changing stereo to mono halves the uncompressed sample count if other settings stay fixed, but a true downmix must combine signals appropriately and avoid phase cancellation or clipping. Simply deleting one channel can lose content.
3. Containers, codecs and compression modes
| Name | Typical encoding/storage | Character | Common application |
|---|---|---|---|
| WAV | Often linear PCM; may contain other encodings | Usually large and editable | Recording, editing and interchange |
| AIFF | Often PCM | Common in some professional workflows | Production and archive |
| FLAC | Lossless compressed audio | Exact decoded samples | Music archive, distribution |
| ALAC | Lossless compressed audio | Exact decoded samples | Apple-friendly lossless libraries |
| MP3 | Lossy audio coding | Compact, broad support | Music and general distribution |
| AAC | Lossy coding, often MP4/M4A | Efficient delivery | Streaming and mobile devices |
| Opus | Lossy coding; Ogg/WebM/other contexts | Efficient speech/music at varied bitrates | Communications, web audio |
WAV is a container and “WAV = uncompressed” is not universally true. A .m4a extension describes a container convention; the audio inside may be AAC or ALAC. A .webm can contain audio with Opus or Vorbis. Confirm the embedded codec rather than assuming it from a filename.
Lossless codecs reconstruct their input samples exactly; they cannot improve a damaged input. Converting a low-bitrate MP3 to FLAC may increase file size but does not recover missing transient detail or frequency content.
4. Lossy audio and psychoacoustics
MP3, AAC and Opus reduce data by exploiting characteristics of auditory perception and sound structure. They are not interchangeable at the same numeric bitrate because their algorithms differ. Perceptual artifacts can include smeared transients, pre-echo, warbling, stereo-image changes and loss of quiet details behind louder sounds.
At a given bitrate, a clean solo voice may be easier to encode acceptably than dense music with cymbals and crowd noise. A 128 kbps AAC and a 128 kbps MP3 are not guaranteed to sound identical. Listen to representative test clips, especially percussive details, sibilance and quiet ambience, rather than relying on a generic percentage quality claim.
Constant bitrate (CBR) targets a consistent rate; variable bitrate (VBR) changes allocation based on signal complexity. Real outputs can differ from simple theoretical estimates because of encoder padding, frames, headers and duration handling.
5. Loudness, peaks and clipping
Peak level describes the maximum sample or reconstructed waveform amplitude, while perceived loudness depends on time, spectrum and hearing. Turning a waveform up to 0 dBFS does not ensure consistent perceived volume. Loudness measurements such as integrated LUFS help compare programs, but targets depend on platform, genre and editorial requirements; there is no single universally correct loudness value.
Clipping occurs when values exceed the allowed range and are flattened or distorted. A file can have no digital clipping yet sound over-compressed from aggressive limiting. True-peak overs may occur between stored samples after digital-to-analog reconstruction or lossy encoding. Leave appropriate headroom for the use case and check after transcoding.
Our Audio Equalizer changes frequency balance; it does not automatically master a recording. EQ can accentuate noise or cause peaks to exceed safe levels. Compare the processed and original audio at matched perceived loudness so louder is not mistaken for better.
6. What happens when you resample or change bit depth?
A proper sample-rate converter estimates the waveform and generates a new set of sample positions. High-quality algorithms filter frequencies that cannot be represented at the new output rate, which helps avoid aliasing. A naive algorithm that simply drops samples or repeats them can produce audible errors.
Reducing bit depth introduces quantization noise. Dither is intentionally added low-level noise that can reduce correlated distortion when truncating precision in suitable PCM workflows. Dither is not a reason to repeatedly convert already-lossy audio. For routine consumer listening, correctly encoded 16-bit PCM often provides ample dynamic range, but professional workflows may retain 24-bit recording and processing headroom.
When preparing files for video editing, 48 kHz is a common interchange rate. Music distribution pipelines often encounter 44.1 kHz. Instead of resampling back and forth, choose a project rate and minimize conversions.
7. Trim audio without surprises
Trimming a compressed audio file can mean cutting at frame boundaries or decoding and re-encoding for more precise positions. Lossy codecs may add delay and padding; the output's exact sample count or first audible transient can shift depending on software. For voice clips, listen to the cut edges for removed consonants, pops and unfinished breaths.
- Keep an unmodified copy of the source.
- Open the Audio Trim tool and mark a safe test segment with a clear audible cue.
- Save the clip, then inspect its first and last second and verify the duration in a second player.
- Compare loudness and stereo channels against the source. If you need sample-accurate editing for production, validate the exported result in a dedicated audio editor.
Browser capabilities and export support vary with the device and software. Do not assume all possible output formats are offered simply because a browser can play them.
8. Voice, podcasts, music and archival choice
| Use case | Good starting approach | What to check |
|---|---|---|
| Spoken-word preview | Opus/AAC/MP3 at tested speech quality | Sibilance, breathing, bandwidth, compatibility |
| Music streaming export | AAC or Opus where supported | Transients, stereo image, target platform |
| Multitrack production | PCM WAV/AIFF or agreed lossless interchange | Sample rate, channel labels, bit depth |
| Long-term owned archive | Lossless master with verified backups | Tags, checksums, playback after migration |
| Video dialogue | Audio matching destination container and workflow | Audio/video sync, levels, channels |
| Accessibility transcript | Separate accurately reviewed text + media | Speaker names, timestamps, language |
A high-resolution label does not guarantee good source recording, mastering or meaningful audible improvement. For accessibility and searchability, pair speech audio with a manually checked transcript or captions rather than treating speech-to-text output as infallible.
9. Troubleshooting unexpected sound
Audio plays at the wrong speed: verify actual rate and header interpretation; accidentally changing a declared sample rate without conversion changes playback speed and pitch.
The WAV is enormous: compute the expected PCM payload and inspect sample rate, depth and channels. Lossless FLAC may be useful for storage while keeping exact decoded samples.
MP3 → WAV still sounds bad: decoding does not recreate lost information. Use the original lossless recording if available.
Left/right sound disappears after conversion: check channel mapping and downmixing. Do not discard channels without inspection.
The end of a trim contains a click: waveform discontinuities at an edit boundary can produce a click; a short fade or adjusted cut point may help.
Speech transcription misses words: check noise, reverb, language selection, speaker overlap and verification. Do not assume AI transcripts are accurate enough for legal or medical purposes without human review.
Export sounds louder but harsher: compare at matched loudness and inspect peaks/limiting or EQ changes.
10. A practical delivery and archiving checklist
- Record the source file's codec, sample rate, channel count and actual duration.
- Decide whether this is an editable master, lossless archive or lossy delivery copy.
- Avoid “upconverting” MP3 to lossless as a purported quality upgrade.
- Set an appropriate project sample rate and minimize repeated resampling.
- Listen to the beginning, middle and end of outputs; inspect channel mapping and sync.
- Check loudness, true peaks and clipping appropriately for the destination.
- Preserve masters and verify archival copies with checksums.
Try it now: Audio Trim · Audio Equalizer · AI Transcriber · Checksum Tutorial.
Technical references
- MDN: Audio codec guide — codecs, containers and compatibility.
- MDN: Media formats and containers — stream packaging and browser support.
- IETF RFC 6716: Definition of the Opus audio codec — normative codec reference.