中文文档
SpecView - Audio Spectrogram Viewer for VS Code
A VS Code extension for viewing audio spectrograms with playback, auto-grouping for A/B comparison, and ML-powered audio classification.
Based on SpecView by Heinrich.
Features
Spectrogram Visualization
Open any audio file to see its spectrogram rendered with a hot-metal colormap. Frequency labels and time rulers are displayed alongside the spectrogram for easy reference.
Supported formats: WAV, MP3, OGG, FLAC, M4A, AAC, WebM, WMA, AIFF, Opus.
Audio Playback
- Play / Pause / Stop controls in the toolbar
- Click on spectrogram to seek to any position (preserves playback state)
- Volume control slider
- Real-time playhead tracking with time display
- Auto-pause: Playback automatically pauses when you switch to another VS Code panel or tab
Time-Domain Waveform Display
Toggle the waveform view via the Waveform checkbox in the toolbar to display the time-domain amplitude envelope above each spectrogram.
- Amplitude labels (-1.0, 0, 1.0) are displayed on the left Y-axis
- The waveform canvas is time-aligned with the spectrogram; click, zoom, and seek operations work on both simultaneously
- Shows amplitude envelope as min/max vertical bars per pixel column
- Synced playhead with the spectrogram view
- Ctrl+wheel and Shift+drag work on both the waveform and spectrogram
Time Axis Zoom
Zoom into any region of the spectrogram for detailed inspection:
- Ctrl + Mouse Wheel: Zoom in/out centered on mouse position
- Mouse Wheel (when zoomed): Horizontal scroll to pan the visible region
- Shift + Drag: Select a time region to zoom into (box selection)
- Toolbar + / – / Fit: Zoom in, zoom out, or reset to full view
- Keyboard:
Shift+Up (zoom in), Shift+Down (zoom out), Shift+Left (fit all)
- Group sync: All tracks in a diff group zoom and scroll together
- Playhead hides automatically when outside the visible region
- STFT is computed once and cached; zoom/scroll operations are instant
Auto-Grouping for A/B Comparison
Files with matching base names and recognized tag suffixes are automatically grouped side-by-side for easy comparison. Use Shift + Space to instantly switch between tracks in a group while maintaining playback position.
The following tags trigger automatic grouping when used as a suffix (e.g., file1_pred.wav) or prefix (e.g., pred_file1.wav) with _ or - separators:
| Category |
Tags |
| Original / Reference |
orig, original, ref, reference, gt, ground_truth, target |
| Generated / Predicted |
pred, predicted, gen, generated, synth, synthesized, output |
| Processing |
recon, reconstructed, enhanced, denoised, clean, noisy |
| Input / Source |
src, source, input, baseline, model |
| Version / Label |
v1, v2, v3, v4, a, b, c, d |
Grouping Examples
| Files |
Grouped? |
Reason |
song_pred.wav + song_orig.wav |
Yes |
stem=song, tags=pred+orig |
song.wav + song_pred.wav |
Yes |
stem=song, tags=(empty)+pred |
pred-song.wav + orig-song.wav |
Yes |
prefix mode, stem=song |
file1_ground_truth.wav + file1.wav |
Yes |
longest match: ground_truth |
/exp1/file1.wav + /exp2/file1.wav |
Yes |
same-name different-dir, tags=exp1+exp2 |
my_song.wav + my_voice.wav |
No |
song and voice not in tag list |
file1_xxx.wav |
No |
xxx not in tag list |
Grouping Rules
- Tags are matched case-insensitively, longest match first
- Both suffix (
stem_tag) and prefix (tag_stem) patterns are supported
- A group forms when 2+ files share the same stem with 2+ distinct tags (empty tag counts as one)
- Same-name files from different directories are auto-grouped; the tag displays as the parent folder name
- Duplicate files (same path) are automatically skipped
ML Audio Classification
Click Analyze to run on-device audio classification using the CED-tiny ONNX model, which recognizes 527 AudioSet sound categories.
Three levels of analysis:
| Button |
Location |
Scope |
| Analyze All |
Toolbar |
All loaded tracks |
| Analyze Group |
Group card header |
All tracks in the group |
| Analyze |
Individual track / lane |
Single track |
The model is downloaded once on first use (~20MB) and cached for subsequent analyses. Detection results are displayed as colored tags below each spectrogram, showing detected sound categories, time ranges, and confidence levels.
Cross-Directory Support
When comparing files from different directories, display names automatically show the relative path from their common parent directory for clear identification.
Example: files from /project/exp1/file1.wav and /project/exp2/file1_pred.wav display as exp1/file1.wav and exp2/file1_pred.wav.
Files with the same name from different directories are automatically grouped, with the parent folder name displayed as the tag.
Usage
Opening Files
| Method |
Description |
| Double-click |
Click an audio file in Explorer to open directly in SpecView |
| Right-click |
Select multiple audio files → "Open with SpecView" → opens in one panel |
| Command Palette |
Ctrl+Shift+P → "SpecView: Open Audio File" → file picker |
| Click drop zone |
Click "Click to add audio files" area → file picker (defaults to last opened directory) |
Track Management
- Delete track: Press
Delete key or click the × button on the card header to remove a standalone track
- Remove lane from group: Click the × button on any lane label within a diff group to remove just that lane
- If only 2 lanes remain, removing one converts the other to a standalone track
- Cards maintain their position in the list when lanes are removed or groups are restructured
- Re-add removed files: Files that have been removed can be re-added normally
Keyboard Shortcuts
| Key |
Action |
Space |
Play / Pause |
Shift + Space |
Switch lane in diff group (A/B comparison) |
Escape |
Stop playback and reset position |
← / → |
Seek backward / forward 2 seconds |
↑ / ↓ |
Switch to previous / next card (track or group) |
Shift + ↑ |
Zoom in (time axis) |
Shift + ↓ |
Zoom out (time axis) |
Shift + ← |
Reset zoom (fit all) |
Delete |
Remove active track (or remove lane from group) |
Requirements
- VS Code 1.85.0 or later
- Internet connection for first-time ML model download (~20MB from HuggingFace)
License
Apache-2.0
Credits
- Based on SpecView by Heinrich
- Audio classification powered by CED-tiny (ONNX Runtime Web)
