Quickstart

1) Launch the GUI

• pixi run gui

2) Create or open a project

From the GUI dashboard:

• Create a project with sample rate, channel mode (mono/stereo), bit depth (8/16/24/32), and file format (wav/flac/ogg), or
• Open an existing project from anywhere on disk, or
• Re-open from recent projects.

3) Manage and explore project files

In the Manage and Explore Files tab:

• Import audio files — originals go to data/raw/, normalized copies go to data/normalized/ automatically
• Optionally set a filename prefix during import to avoid collisions across sources with the same filename
• Apply one label to all files uploaded together
• Rename files (also renames the matching raw counterpart)
• Label files with one or more comma-separated labels (e.g., bab-f1, studio) for organization
• Rename labels to apply a new label to all files with the old label (multi-label files: only the target label is replaced)
• Bulk label unlabeled files — apply a single label to all files with no labels at once
• Filter by label to view only files with a specific tag; use (None) to show only unlabeled files
• Remove files (also removes matching raw counterpart and spectrogram artifacts)
• Play files inline
• View precomputed spectrograms (generated at import time from the normalized file)

4) Generate babble speech (mix multiple talkers)

In the Babble tab:

• Label talker-group files as bab-f1, bab-f2, bab-m1, etc., or assign that label during import to a whole batch
• Set the total number of talker groups to mix
• Optionally set female and male counts separately
• Set intended per-talker length (seconds)
• Normalize each source file to the same RMS, concatenate files for each talker, and optionally peak-normalize the output
• Play and download the mixed babble output
• Add the generated babble back into the project as a derivative labeled bab-t#, where # is the number of talkers used

Or from CLI:

# Label talker groups
pixi run triton files label my-project speaker_f1.wav "bab-f1"
pixi run triton files label my-project speaker_f2.wav "bab-f2"
pixi run triton files label my-project speaker_m1.wav "bab-m1"

# Mix 3 talker groups with RMS normalization
pixi run triton babble mix my-project \
  --num-talkers 3 \
  --intended-length-seconds 30 \
  --output-path outputs/babble.wav

5) Mix speech with noise in GUI

In the Mix tab:

• Upload speech and noise
• Set target SNR
• Play and download the mixed output

6) Mix speech with noise in CLI (path-based)

• pixi run triton mix path/to/speech.wav path/to/noise.wav --snr-db -5

7) Ingest from RSS

• pixi run triton ingest rss --feed https://feeds.megaphone.fm/the-moth --output-dir data/moth --limit 10

8) Transcribe locally (optional feature)

Requires the transcribe feature (see Install):

• pixi run --feature transcribe triton transcribe local data/moth/example.mp3 --output-dir outputs/transcripts --model tiny

9) Apply degradations

• pixi run triton degrade vocode data/speech --vocoder-type noise --n-bands 8

10) Convert formats

• pixi run triton convert resample data/audio --target-sr 16000
• pixi run triton convert mono data/stereo

11) Run a pipeline matrix (parameter sweep)

Once you've defined a pipeline in your project:

# Generate a CSV of file × parameter combinations
pixi run triton matrix generate my-project my-pipeline matrix.csv \n  --param '0.target_peak=0.5,0.8' \n  --file normalized/file1.wav \n  --file normalized/file2.wav

# Run all rows in the matrix
pixi run triton matrix run my-project my-pipeline matrix.csv

Each row produces isolated outputs so results are easy to compare and aggregate. See Pipeline Matrix for details.

File Labeling & Organization

Organize your project files with labels for easier workflows:

# Label files
pixi run triton files label my-project speaker_f.wav "bab-f1"
pixi run triton files label my-project background.wav "noise"

# List files by label
pixi run triton files list my-project --label "bab-f1"

# View all labels
pixi run triton files show-labels my-project

See Files for more details.

Babble Speech Workflow

Generate babble by mixing labeled talker groups with RMS normalization:

# Mix talker groups
pixi run triton babble mix my-project \
  --num-talkers 3 \
  --intended-length-seconds 30 \
  --output-path outputs/babble_3talkers.wav

# Or specify female and male counts explicitly
pixi run triton babble mix my-project \
  --num-talkers 4 \
  --num-female-talkers 2 \
  --num-male-talkers 2 \
  --intended-length-seconds 45 \
  --output-path outputs/babble_4talkers.wav

# Add noise to the babble at a target SNR
pixi run triton mix outputs/babble.wav noise.wav --snr-db -5

If you save a generated babble back into the project from the GUI, Triton stores a provenance sidecar with the source files and babble-generation parameters so the derivative can be traced later.

See Babble for full command reference.

Notes

• Project lifecycle and storage logic now live in triton.core.project so GUI and future CLI project commands can share the same implementation.
• Current CLI commands remain path-oriented; project-first CLI commands can be layered on top of the shared core module.
• Pipeline runs are grouped by run folder, and each step writes to its own step folder under data/derived/pipelines/....
• Generated artifacts include sidecar provenance JSON files (<artifact>.<suffix>.json) to record source and action history.
• Imported files are automatically normalized to the project spec (sample rate, channels, bit depth, file format) and saved to data/normalized/. Spectrogram artifacts are generated from the normalized file using project defaults in triton.toml ([spectrogram]).
• Pipeline Matrix allows reproducible batch processing by combining multiple files and parameter combinations in a single run.
• File labels persist in metadata/file_labels.json and work across CLI and GUI interfaces.
• Babble generation in GUI and CLI uses the same shared core function in triton.degrade.noise_generator.