Crossfade Concatenator — User Guide

What this does

Advanced Concatenate with Crossfade v1.4 creates a new sequence from one or more selected Sound objects. Each source can contribute its whole duration or one or more fixed/random excerpts. The resulting chunks can remain in extraction order, be uniformly shuffled without replacement, or be sampled with replacement. Adjacent chunks are joined by a manual overlap-add crossfade using one of five selectable curves.

The processor supports mono and stereo material. Mixed sample rates are resampled to the sampling frequency of the first selected Sound. Mixed mono/stereo inputs are converted to the channel count of the first extracted chunk. Material with more than two channels is rejected.

Quick start

1. Select at least one Sound object in Praat.
2. Run Concatenate_with_crossfade_v1.4.praat.
3. Choose a preset, or leave Custom and set chunking, order, crossfade, overlap, and dynamics manually.
4. Enable Draw_visualization if you want the result waveform, segment map, dynamics envelope, file legend, and summary.
5. Click OK. The final Sound is named concat_crossfade_<PresetName>.

Parameters

Chunk extraction and order

Crossfade and overlap

Dynamics and output

Factory presets

Chunk extraction, compatibility, and playback order

Extraction

Fixed and random chunks are taken from a uniformly random valid offset inside the source Sound. The code uses each Sound object's actual start and end times, so sources with non-zero time origins are handled correctly. A requested chunk longer than its source is shortened to the full available duration.

Sample rate and channels

The sampling frequency of the first selected Sound is the target. Any extracted chunk with a different rate is resampled to that rate with Praat's Resample: sr, 50.

Only mono and stereo chunks are accepted. The channel count of the first extracted chunk becomes the target channel format: stereo chunks are converted to mono when the target is mono; mono chunks are duplicated into stereo when the target is stereo.

Order

Crossfade curves and overlap rules

For normalized overlap position u from 0 to 1, the outgoing and incoming chunks are multiplied by complementary curves, then summed by manual overlap-add.

Actual overlap duration

The selected overlap mode first produces a requested overlap. The script then caps it to 90% of both neighboring chunks:

capacity = min(0.9 × incomingDuration, 0.9 × outgoingDuration)
overlap = min(requestedOverlap, capacity)
minimum = min(0.01 s, capacity)
overlap = max(overlap, minimum)

Therefore a join never consumes an entire neighboring chunk. When both chunks permit it, the overlap is at least 10 ms; for extremely short chunks, the available 90% capacity takes precedence.

Dynamics modes

Let d = Dynamics_depth_percent / 100 and minAmp = 1-d. A depth of 0% gives unity gain; 100% allows the chosen envelope to reach zero where its shape reaches its minimum.

Processing pipeline

1. Store the selected Sound object IDs and use the first Sound's sampling frequency as the target.
2. Apply the selected preset, then sanitize inverted chunk/overlap ranges, dynamics depth, and peak target.
3. Extract Chunks_per_file chunks from each source.
4. Resample mismatched chunks to the first Sound's sampling frequency.
5. Reject channel counts other than 1 or 2, then convert all chunks to the first extracted chunk's mono/stereo format.
6. Build sequential, shuffled, or with-replacement playback order.
7. For Random/Terraced dynamics, assign the per-position gains before rendering.
8. Starting from the first playback chunk, calculate each overlap, apply exactly one fade-out/fade-in pair, and sum the buffers with manual overlap-add.
9. Apply any continuous whole-output dynamics envelope.
10. If the result is non-silent, scale its absolute peak to Scale_peak; if completely silent, skip peak scaling.
11. Rename to concat_crossfade_<PresetName>, compute output RMS, optionally draw, clean temporary chunks/control objects, and optionally play.

Visualization

When Draw_visualization is enabled, v1.4 draws a vertically stacked 8-inch report:

• Header: preset, source count, chunk count, crossfade type, and dynamics mode.
• Panel A — Result waveform: final output with dashed segment-start markers.
• Panel B — Segment map: time-aligned colored rectangles; color identifies source file and the number shows playback position when space permits.
• Panel C — Dynamics envelope: flat unity line for None; analytic curve for continuous modes; for Random/Terraced, a parallel one-channel control signal is assembled through the same overlap-add crossfades as the audio.
• Panel D — File color legend: one swatch per selected source, with long names truncated for display.
• Panel E — Summary: preset/order, chunk mode, crossfade and overlap settings, dynamics, output name/duration, total overlap, output RMS, and sample rate.

Limits and edge cases

• Mono/stereo only: any extracted chunk with another channel count stops the script.
• Target format comes from the first source/chunk: sample rate follows the first selected Sound; mono/stereo format follows the first extracted chunk.
• Randomness is not seeded in the form: random chunk offsets, durations, orders, overlaps, and segment gains may change between runs.
• Overlap is constrained: every join is capped at 90% of each neighbor, regardless of a larger percentage/fixed/random request.
• Scale_peak above 1 is clamped: this avoids intentionally creating a post-normalization peak above digital full scale.
• Silent output: final peak normalization is skipped instead of calling Scale peak on an all-zero Sound.
• Random with repeats: the number of output positions remains totalChunks, but the underlying extracted chunks are sampled with replacement.