Keep an uncompressed PCM WAV when you need a durable editing source, and make an MP3 copy when upload size is the immediate constraint. A carefully checked MP3 can be a practical transcription input, but the extension does not guarantee intelligible speech. Preserve the best original before making the smaller version so that a disputed word can still be checked against it.
WAV describes more than one kind of file
Audacity's export-format guide distinguishes signed 16-bit PCM WAV, signed 24-bit PCM WAV and 32-bit float WAV. It also documents compressed audio inside WAV, including GSM and custom codec combinations. Calling a file WAV is therefore not enough to establish that it contains uncompressed PCM. Inspect the export setting, not just the filename.
For an editing source, the useful property of PCM is avoiding another lossy encoding stage. That does not turn a poor recording into a good one. If the source was already an MP3, decoding it into WAV cannot reconstruct information discarded during the original compression. The larger new file is not evidence that previously missing words have returned.
Bit depth, sample rate and channel count affect PCM file size. They answer different questions: how each sample is represented, how many samples are stored per second, and how many channels are kept.
Audacity also lists FLAC as compressed but lossless. That shows why compressed and lossy are not synonyms. A destination's supported formats still matter; being lossless does not by itself make a file acceptable to every transcription uploader.
MP3 settings control the transfer trade-off
Audacity offers variable, average and constant bitrate options. Variable bitrate changes with the material, so the exact file size cannot be predicted in advance. Constant bitrate makes a useful size estimate possible. That is valuable for a fixed upload ceiling, but it is not a claim that constant bitrate always gives the best sound.
The manual's spoken-word guidance describes 64 kbit/s constant bitrate mono as a small-file option and 96 kbit/s mono as a higher-quality voice option. These are listening-oriented recommendations, not transcription accuracy benchmarks. Use them as candidate settings for a short comparison, not as a universal rule for a quiet interview or difficult terminology.
The same guide recommends mixing identical left and right spoken-word channels to mono before constant-bitrate export. The condition matters: two identical channels are different from one speaker isolated on the left and another on the right. Listen to each channel before collapsing them. Retain the separate source if it is useful for later correction or editing.
Repeated MP3 export is another avoidable compromise. Audacity warns that importing, editing and re-exporting MP3 introduces another lossy encoding stage. Work from the original in a lossless editing workflow, then create the delivery MP3 when the edits are finished.
One hour of speech, calculated openly
For PCM, audio-data bytes equal sample rate multiplied by bit depth, channel count and duration, divided by eight. For constant-bitrate MP3, multiply bits per second by duration and divide by eight. The following hypothetical examples use 3,600 seconds and decimal MB, meaning one million bytes. Container headers and metadata add a little more.
| One-hour file | Calculation | Approximate audio data |
|---|---|---|
| PCM WAV, 16 kHz, 16-bit, mono | 16,000 × 16 × 1 × 3,600 ÷ 8 | 115.2 MB |
| PCM WAV, 44.1 kHz, 16-bit, mono | 44,100 × 16 × 1 × 3,600 ÷ 8 | 317.5 MB |
| MP3, 64 kbit/s constant bitrate | 64,000 × 3,600 ÷ 8 | 28.8 MB |
| MP3, 96 kbit/s constant bitrate | 96,000 × 3,600 ÷ 8 | 43.2 MB |
| MP3, 128 kbit/s constant bitrate | 128,000 × 3,600 ÷ 8 | 57.6 MB |
These calculations do not establish equal audio quality or equal transcripts. They show why changing from WAV to MP3 is not enough information: a 128 kbit/s copy can still exceed a 50 MB upload limit, while the two lower-bitrate examples leave room beneath it. Inspect the real exported file's size before uploading.
A hypothetical student needs to upload a one-hour lecture and retains a 115.2 MB PCM source. A 96 kbit/s copy is a reasonable candidate to check because its calculated audio data is 43.2 MB. If important words become less clear, preserving intelligibility takes priority over forcing that setting. Use a different acceptable transfer arrangement rather than overwriting the source.
Ten minutes to check the actual export
Choose a short section containing your hardest material: a quiet sentence, a name and a technical expression. The aim is to expose a bad transfer choice before converting an entire archive.
- Spend two minutes recording the original's duration, channels and encoding. Save an untouched copy with a clear name such as source-original, separate from upload-copy.
- Use two minutes to export the selected passage with your proposed MP3 settings. Note whether the bitrate is constant or variable. A format label alone is not enough to repeat the result.
- Spend three minutes listening to the original and exported passage through the same headphones at a comparable volume. Check the difficult words and the start and end of the selection.
- Use three minutes to open the exported file in the destination workflow, if processing is permitted. Verify the resulting text against the source, then record whether the size saving is worth the extra correction work.
Processing may take longer than this rehearsal budget. Wait for completion rather than interpreting a pending result as a failed format. Once a setting passes your small check, export the full recording from the preserved source and verify its duration again.
Do not confuse a convenient copy with a master
Audacity notes that MP3 encoding introduces a short delay at the beginning. It writes information that compatible applications can use to remove that added silence, but not every player supports it. If precise alignment with video or separate tracks matters, the manual recommends formats such as WAV or FLAC instead. A file that sounds fine alone may still need an alignment check.
Renaming a filename from MP3 to WAV is not conversion. Likewise, exporting to a higher bitrate cannot recreate information already removed. The practical safeguard is simple: keep the best available recording and identify derived copies clearly. That lets you revisit an uncertain passage without guessing which version was compressed several times.
There is no measured transcription winner here. The source may contain an unclear word before either export is made. More megabytes cannot demonstrate that the word was captured, and a fluent transcript cannot prove that it was recognised correctly. Resolve important uncertainty by listening to the original.
The 50 MB decision in Notibo
Notibo accepts audio or video uploads up to 50 MB. In the calculated one-hour examples, the PCM WAV and 128 kbit/s MP3 exceed that ceiling; the 64 and 96 kbit/s examples fit below it. Leave room for the actual file structure and check the exported size rather than treating the calculation as an upload guarantee.
Notibo produces a transcript with structured notes, including a summary and key points. It processes in the cloud, with EU storage and US processors, rather than on your device. Select one main language; automatic detection finds the dominant language, while switching languages can cause errors.
Use the smaller copy for transfer and the original for verification. Notibo does not promise transcript timestamps or speaker labels, so keep your own connection between the source and its notes. The sensible outcome is a usable transcript plus an intact source, not the smallest file at any cost.
Related comparisons
- Auphonic vs Audacity: prepare speech for transcription
- Zoom H1essential vs a phone: do you need a recorder?
This is a documentation-based comparison by Notibo, checked on 20 October 2026, not a measured accuracy benchmark. The examples are suggested trials, not reported test results.