What Bit Rate Does Spoken-Word Audio Need?

Spoken-word files — audiobooks, podcasts, lecture recordings, oral history — are routinely a fraction of the size of music files of the same length. The reasons are structural, and they are all visible in the bit-rate arithmetic.

Start with the uncompressed number

For uncompressed PCM, the bit rate is fixed by three values:

sample rate × bit depth × channels = bits per second

Sample rate is samples captured per second; the FADGI glossary entry ties it to the Nyquist–Shannon basis. Bit depth is the word length of each sample, which is what sets the encoded dynamic range — see the companion bit depth entry.

So 44,100 × 16 × 2 = 1,411,200 bits per second, about 1,411 kbps — the CD figure. Drop to one channel and it halves to about 706 kbps, before any compression at all. That single change is the largest one available to spoken word, because narration carries no stereo image worth preserving.

The full walkthrough is in how to calculate audio bit rate and file size.

Why speech compresses further

Three properties of speech recordings work in favor of small files:

Mono source. One narrator, one microphone, one channel.
Limited content bandwidth. The energy that carries intelligibility sits well below the top of the audible band, so encoders have less high-frequency detail to spend bits on.
Sparse texture. One voice at a time, with pauses, is far easier for a perceptual encoder to model than a dense mix.

Lossy codecs exploit all three. Spoken-word distribution commonly lands in the tens of kbps mono, where music distribution sits in the low hundreds of kbps stereo. Distribution catalogs differ, so treat any particular figure as a platform choice rather than a technical constant.

Constant versus variable bit rate

A constant bit rate (CBR) file spends the same bits per second throughout, which makes file size trivially predictable: bit rate in kbps ÷ 8 gives kilobytes per second. One hour at 64 kbps is about 28.8 MB.

A variable bit rate (VBR) file spends more bits on difficult passages and fewer on silence and steady speech. For narration with frequent pauses, VBR usually produces a smaller file at the same perceived quality, but the size is only known after encoding. Players report VBR files by their average bit rate, which is why a stated figure and a measured file size can disagree.

Archive copy versus distribution copy

These are different files with different jobs, and conflating them is the common mistake.

The archive copy is the one you keep. The Library of Congress Recommended Formats Statement for audio works points to uncompressed PCM in a WAVE container at 44.1 kHz/16-bit or higher for preservation. Uncompressed, so nothing is discarded; format-stable, so it opens in twenty years.

The distribution copy is the one listeners download. It is derived from the archive copy, encoded lossily, and disposable — because it can be regenerated whenever a target format changes.

Between the two sits lossless compression. FLAC reduces file size with no change to the decoded samples, and its behavior is documented by the nonprofit maintainer in the Xiph.Org FLAC documentation. For speech, the savings are real but modest compared to lossy encoding, because lossless compression cannot discard anything. See is FLAC really lossless? for what "lossless" does and does not guarantee.

Practical consequences

Re-encoding a lossy file into another lossy format compounds the losses. Go back to the archive copy.
Recording narration in stereo doubles storage for no gain unless the source is genuinely multi-channel.
A high sample rate on a voice recording raises the bit rate without adding intelligibility; is 16-bit / 44.1 kHz good quality? covers why the baseline is usually sufficient.