Chapter 2 Key Takeaways

What you should walk away with

  • Digital audio is not stairsteps. For any signal containing only frequencies below half the sample rate, the samples capture it completely — they define exactly one band-limited wave, and the reconstruction filter rebuilds precisely that wave, smooth and stepless. The blocky zoom in your DAW is a drawing convenience, not the signal. This is the chapter's threshold concept; once it clicks, half the audiophile internet stops making sense.

  • Sample rate sets the captured frequency band — nothing else. The Nyquist frequency (sample rate ÷ 2) is the honest ceiling: 22,050 Hz at 44.1 kHz, 24,000 Hz at 48 kHz. Both clear human hearing entirely, which is why no blind test reliably tells them apart.

  • Aliasing is what happens above the ceiling. Content past Nyquist doesn't vanish — it folds back into the audible band as false, inharmonically related tones (a 30 kHz intruder at 44.1 kHz lands at 14.1 kHz). The anti-aliasing filter must act before conversion, because a recorded alias is indistinguishable from real audio forever after. Today you'll mostly meet aliasing inside nonlinear plugins — that's what their oversampling/HQ switches are for.

  • Bit depth is a noise floor, not a quality dial. Each bit buys ~6 dB of dynamic range: 16-bit ≈ 96 dB (already below audibility at any sane playback level), 24-bit ≈ 144 dB theoretical. Loud signals are captured identically at either depth.

  • Track at 24-bit anyway — it's headroom insurance. The deep floor lets you record with cowardly, generous margins (peaks around -18 to -10 dBFS) and boost later for free. Sloppy gain gets forgiven; quiet sources stay clean; the cost is 50% more disk, which is nothing.

  • dBFS is the ruler with zero at the top. It measures distance to the digital ceiling; dB SPL measures sound in air from the bottom up. No fixed conversion exists between them, and conflating them is the root of half the level confusion in home studios.

  • 0 dBFS is a cliff, and capture clipping is permanent. Above full scale there are no numbers; peaks arrive sheared flat, spraying harsh odd harmonics (and possibly aliases). Inside the DAW's float math an over is usually survivable — at the converters and in printed files, it's an amputation. Never clip a take, never clip a final file.

  • Dither once, at the end. Reducing to 16-bit turns rounding error into music-correlated grunge on fades; a whisper of noise added first dissolves it into harmless hiss. One checkbox, at the final 16-bit bounce only — never stacked, never agonized over.

  • The 192 kHz folklore fails at Nyquist. High rates annex ultrasonic territory your ears, mics, and speakers never visit — and can even backfire on playback gear. The honest exceptions: sound-design capture you'll slow way down, and alias-safety for heavy processing (mostly handled by plugins oversampling internally). Default to 48 kHz / 24-bit and spend your attention where listeners can hear it.

  • Buffer size is the tradeoff you can feel. Small buffers = low latency but fragile CPU deadlines; big buffers = stability with audible monitoring delay. Track at 64–128 samples, mix at 512–1024, use direct monitoring as the escape hatch. Every decision is a tradeoff; this one you renegotiate weekly, calmly.

  • Lossy is an endpoint, never a source. MP3/AAC store a perceptual bet, not the waveform; cymbals, applause, and reverb tails pay first when the bet loses. Work and archive lossless (WAV/AIFF/FLAC), encode deliverables fresh from the master, leave lossy encodes ~1 dB of peak headroom, and never re-encode lossy from lossy.

  • One archive master, many derived deliverables. A 24-bit WAV at session rate with honest headroom is the reference truth every other version descends from. Platforms transcode whatever you send — feed the robots the lossless original.

Remember this

  • The samples aren't a sketch of the wave — they're a perfect set of instructions for rebuilding it.
  • Sample rate is how often; bit depth is how far down the hiss lives; dBFS is how close to the cliff.
  • Nobody ever heard your sample rate. Everybody hears a clipped take.
  • Settings are chosen once, on purpose, in daylight — never at 1 a.m. on a deadline.

🎚️ "Static Bloom" status

The project exists. Jaylen created the session that will carry the whole book — 48 kHz / 24-bit, decision documented, buffer at a neutral 256, folder skeleton built (Session / Audio / Samples / Refs / Bounces / Exports / Notes), references imported, samples copied in rather than referenced, saved as v0.1 under a working title. (It won't earn the name "Static Bloom" for months — the name is waiting inside a synth patch that doesn't exist yet.) Aisha, in parallel, templated her podcast sessions and taped the three numbers to her monitor. Your track: you should now have your own project at 48/24 (or 44.1/24 with the reason written down), the folder tree around it, a saved template, and your settings on something physical. If you skipped it, it's a 30-minute assignment — do it before Chapter 3, which needs a session to wire up.

Pointing forward

Chapter 3 follows the signal itself — from sound wave to speaker through every box and cable you own — and turns "my recording sounds wrong somewhere" into a chain of specific, checkable stages. The ideas you just built compound later at known addresses: headroom and dBFS become the gain-staging workflow of Chapter 21; the clipping cliff's gentle, musical cousins arrive as tools in Chapter 26; the dither moment and master-delivery specs land in Chapters 31–32; and the loudness system that quietly judged Aisha's Episode 47 gets its full reckoning — LUFS, true peak, normalization, and the end of the loudness war — in Chapter 33. Exercises and quiz first, though: the blind tests in this chapter's exercises are where these ideas stop being things you read and start being things you've verified.