Audio Formats and Bitrates: Choosing Settings That Are Not Guesswork

Bitrate, sample rate and bit depth

Bitrate

How many bits per second the compressed file spends describing the audio. This is the main quality-versus-size control in any lossy format and the one you should normally be adjusting. 128 kbps means roughly one megabyte per minute.

Sample rate

How many times per second the waveform was measured, in Hz. It sets the highest frequency that can be represented, which is half the sample rate. 44.1 kHz captures up to about 22 kHz, comfortably above human hearing. Raising it beyond 48 kHz does not improve what you can hear; it mostly matters during production, not delivery.

Bit depth

How precisely each sample is measured, which determines the dynamic range between the quietest and loudest representable sound. This applies to uncompressed formats such as WAV. Lossy codecs do not have a bit depth setting at all, so if a tool offers you one alongside an MP3 export it is describing the intermediate stage, not the output.

The formats worth knowing

Opus is technically the strongest option at almost every bitrate, but it is not universally supported by consumer hardware and podcast platforms. MP3 remains the safe answer whenever you do not control what will play the file.

Bitrates that make sense

Speech is far more compressible than music because it occupies a narrow frequency range and has long near-silent gaps. Encoding an interview at 320 kbps stereo produces a file four times larger than necessary that sounds no better, and mono alone halves it with no loss for a single microphone.

Variable bitrate encoding, where available, spends more bits on complex passages and fewer on simple ones. For a given average size it generally sounds better than a constant bitrate. Use it unless something in your delivery chain specifically requires constant.

Extracting audio from video

A video file is a container holding separate video and audio streams. The audio inside an MP4 is usually already AAC, which means it can often be copied out verbatim rather than re-encoded.

• If the existing audio stream is already in a format you can use, copy it. This is instant and lossless because nothing is decoded or re-encoded.
• Only transcode when you need a different format, for example AAC to MP3 for a podcast host that demands it.
• Transcoding from one lossy format to another compounds the losses of both. Where the destination bitrate is similar to the source, the second encode does more harm than the bitrate number suggests.
• Check the channel count. Video soundtracks are sometimes 5.1 surround, and a careless downmix to stereo can leave dialogue, which usually lives in the centre channel, far too quiet.

Loudness, and why your file is quieter than everyone else

Perceived loudness is not the same as peak level. A file can peak at maximum and still sound quiet if its average level is low. Streaming platforms measure the average using the LUFS scale and normalise everything to a target, so exporting louder than the target gains you nothing and simply gets turned down, usually after being limited in ways you did not choose.

• Spoken word content is commonly targeted around -16 LUFS for stereo and -19 for mono.
• Music platforms typically normalise to somewhere between -14 and -16 LUFS.
• Leave a little headroom below maximum peak so that lossy encoding, which can slightly overshoot the original waveform, does not clip.
• Normalise before encoding, not after. Adjusting the level of an already-encoded lossy file means decoding and re-encoding it.