Project and source-file sample rates

When rates differ, the DAW normally resamples during playback. Rendering may use the project rate or another delivery rate.

Sample rate mismatch

If a file is interpreted at the wrong rate without resampling, pitch and speed change:

• 96 kHz file interpreted as 48 kHz → slower, lower pitch
• 44.1 kHz file interpreted as 48 kHz → faster, higher pitch

Sources of latency

Buffer-size trade-off

Buffers that are too small may cause clicks or dropouts.

Buffer size and sample rate

The table shows the duration of one audio processing block. This duration is one component of the overall system latency.

→ For comparison, sound travels approximately one metre through air in 2.9 ms.

Real-time and offline processing

Processing may follow playback in real time or run offline, faster or slower than the audio. Rendering may use either mode.

Bit depth

Clipping concerns the limits of the available amplitude range. Bit depth determines how finely values within that range can be represented.

• in integer PCM, n bits provide 2ⁿ possible sample values
• each sample is assigned to one of these values through quantization
• more available values reduce quantization error and increase the dynamic range

→ Use 24-bit PCM for general recording workflows. 16-bit PCM remains common for some delivery formats, while 32-bit float is used in some recording and processing workflows.

Dither

Dither is low-level noise added before reducing integer PCM bit depth. It helps decorrelate quantization error from the audio signal.

Dither listening example

A 300 Hz tone fading to silence, reduced to 8-bit so the effect is obvious (in practice, dither is applied at the final 16-bit export, where it is far subtler):

1. 16-bit reference, full resolution. ▶ Play
2. 8-bit without dither: the quiet tail distorts and drops to silence. ▶ Play
3. 8-bit with dither: the tail keeps fading, into a faint steady noise. ▶ Play

Audio file formats

• WAV / AIFF: containers commonly used for uncompressed PCM audio
• FLAC / ALAC: lossless compression; the original samples can be recovered
• MP3 / AAC / Ogg Vorbis: lossy compression; smaller files with reduced fidelity

→ Choose a format according to compatibility, file size, and required fidelity.

Technical settings summary

• project sample rate: sets the playback and processing reference
• integer PCM bit depth: determines amplitude resolution and theoretical dynamic range
• recording format: 48 kHz / 24-bit WAV for this course
• buffer size: balances latency against the time available for real-time processing
• dither: apply once when reducing to a lower integer bit depth

Visualization views in DAWs

• Waveform: amplitude over time
• Spectrum: frequency content within an analysis window
• Spectrogram: changing frequency content over time

Waveform view (time domain)

A waveform shows amplitude over time; zooming helps locate transients, edit points, and fades.

Axes: time horizontally · amplitude vertically

Spectrum view (frequency domain)

A spectrum shows frequency content within a short analysis window and helps identify fundamentals, harmonics, and noise.

Axes: frequency horizontally · level vertically

▶ Play displayed excerpt

Reading a spectrogram

A spectrogram is a visual representation of how a sound's spectrum changes.

• Horizontal axis: time
• Vertical axis: frequency
• Color or brightness: magnitude or level

Spectrogram settings

Adjusting FFT and window size affects how detail appears:

• large window: better frequency resolution, less precise timing
• small window: better time resolution, less precise frequency
• common window sizes: 512, 1024, 2048, or 4096 samples

→ Trade-off between time and frequency resolution.

Spectrogram settings

Analysis for sound design

Visual analysis supports critical listening by making spectral and temporal patterns visible.

• locate tonal build-up, hum, hiss, or noise
• compare timbres, transients, and sustained components
• relate visible patterns to audible changes across different time scales

Spectral editing examples

Some applications use the spectrogram as a time-frequency editing surface.

• select a region by time and frequency
• attenuate clicks, hum, noise, or unwanted tones
• adjust selected frequencies, similar to localized EQ

Examples include REAPER spectral edits, iZotope RX, Adobe Audition, and Steinberg SpectraLayers.

Machine-learning source separation

Source-separation models estimate component signals from a mixed recording.

• output categories depend on the model; common music stems include vocals, drums, bass, and other material
• models may process waveforms, time-frequency representations, or both
• separated signals may contain leakage, smearing, and other artifacts
• results can be inspected and used as new editing material

Demucs research paper · Waves Mix Unlock DAW example

Buffer size and voice monitoring

Use the computer’s built-in microphone and wired headphones at a moderate level.

1. select the built-in microphone and wired headphone output
2. arm a track, enable input monitoring, and set the buffer to 64 samples
3. speak while listening to the monitored signal
4. repeat at 1024 samples, compare the delay, then return to the lower setting

Tone sequence and spectrogram

Create a sequence of three 1.5-second sine tones.

1. generate 293.66, 349.23, and 440 Hz using
   JS: Tone Generator or Generate → Tone
2. place the tones one after another in a single file or track
3. inspect the sequence in waveform and spectrogram views
4. locate each pitch and the transitions between them

Which view makes the pitch changes and their timing easier to identify?

Selected references

• Rumsey, Francis, with Tim McCormick. Sound and Recording: Applications and Theory. 8th ed., Routledge, 2021.
• Smith, Julius O. Spectral Audio Signal Processing. W3K Publishing, 2011.

Related: Fundamentals of Sound · Waveforms · Spectral Analysis · Digital Audio