People new to encoding MP3s tend to fall into one trap: assuming 320kbps is always "better" than 128kbps and maxing it out. The result is doubled file size with a barely audible gain. This post sorts the three things that actually decide MP3 quality — sample rate, bit depth, and bitrate — then gives you a cheat sheet for choosing by use case.
The three parameters that matter
Sample rate (Hz) — how many times per second the sound is sampled. CDs use 44100Hz; video commonly uses 48000Hz. It sets the frequency ceiling you can preserve.
Bit depth (bit) — how many bits each sample stores, typically 16bit or 24bit. It sets dynamic range and signal-to-noise.
Bitrate (kbps) — the amount of data encoded per second, i.e. how much information the output file actually holds. Bitrate directly determines MP3 file size and is the knob people reach for most.
Sample rate and bit depth describe how good the raw data (PCM) is; bitrate describes how much survived compression. MP3 is lossy — it's supposed to throw data away — and bitrate is the "how much to drop" dial.
Which bitrate: 128 / 192 / 320
| Level | Perceived quality | Typical use |
|---|---|---|
| 128k | Acceptable; you can hear high-frequency loss | Speech, podcasts, casual sharing |
| 192k | Most people can't tell it from a CD | Everyday music — best value |
| 256–320k | Near-transparent to the human ear | Hi-fi archives, master copies |
Practical takeaway: music starts at 192k; speech is fine at 64–128k. 320k is nearly twice the size of 192k for an improvement most people can't hear.
CBR or VBR?
This is the most-confused concept in MP3.
CBR (constant bitrate) — the whole file is encoded at one fixed rate. Predictable size and best compatibility with old players; but quiet passages pay full bitrate (wasted space) and intense ones can run short.
VBR (variable bitrate) — the encoder allocates bitrate by content complexity, lower during quiet parts and higher during busy ones. Smaller at equivalent quality, and the default for local music libraries; the downside is unpredictable size and rare incompatibility on old hardware.
In FFmpeg the LAME encoder controls these with two different flags:
# CBR: fixed 192k
ffmpeg -i input.wav -c:a libmp3lame -b:a 192k out.mp3
# VBR: a quality setting, -q:a lower = better (0 max)
ffmpeg -i input.wav -c:a libmp3lame -q:a 2 out.mp3
-q:a (LAME's VBR quality scale) runs 0–9, smaller = higher quality / larger file:
| -q:a | Bitrate range | Use |
|---|---|---|
| 0 | 220–260 kbps | Highest quality |
| 2 | ~190 kbps | High-quality sweet spot (recommended) |
| 4 | 140–185 kbps | Everyday |
| 9 | 65–85 kbps | Speech |
Pick CBR or VBR — don't mix them.
Extracting audio or converting to MP3
Our two features split the work:
- Extract Audio: pulls the raw audio track out of a video, defaulting to
-vn -c:a copy— no re-encode, virtually byte-for-byte. - Video to MP3: converts to MP3, lossy via LAME, quality adjustable.
Command-line versions:
# Extract the original track (if the source is mp4/aac, copy yields aac, not mp3)
ffmpeg -i input.mp4 -vn -c:a copy track.m4a
# Convert to MP3 (lossless source → lossy, done in one shot)
ffmpeg -i input.mp4 -vn -c:a libmp3lame -q:a 2 out.mp3
Don't encode lossy twice
MP3 is lossy; once you go from FLAC/WAV to MP3, the discarded data isn't coming back. Three habits to avoid:
- Re-encoding an MP3 into another MP3 (second loss, quality degrades layer by layer)
- "Converting" an MP3 "back" to WAV/FLAC to fake lossless (same truncated data, new container)
- Encoding a lossy m4a to MP3 at 320k (the source is already damaged; max bitrate won't fix it)
Do this instead: keep the one lossless source (FLAC/WAV) as the master, and only generate an MP3 copy when you need to share or save space.
Sample rate and channels: don't touch what works
In almost every case, keep the original sample rate (44.1k / 48k) and stereo. Downsampling the sample rate chops off high frequencies, making vocals and music sound dull; going mono saves half the space but suits speech only.
FAQ
Q: Is 320k always better than 128k? A: Usually to the ear, but past 192k most people can't tell, and only if the source is good. A damaged source won't be saved by 320k; a lossless source is often fine at 192k.
Q: Why isn't a copy-extracted file an mp3? A: copy moves the raw stream as-is; whatever the source codec is (aac / mp3 / flac…), that's what you get. You only get an MP3 from copy if the source is already MP3. Forcing MP3 means re-encoding — which is lossy.