MDCT Sound Compression

Stereo audio compression and decompression using a Mid/Side MDCT pipeline with a desktop GUI and test suite.

Overview

• Goal: Compress stereo audio into a compact binary format and reconstruct it with controlled quality loss.
• Core idea: Convert Left/Right (stereo channels) to Mid/Side (sum/difference), apply MDCT on overlapped blocks, quantize coefficients, and discard the last M coefficients per block.

Algorithm (High-Level)

1. Load audio (WAV/FLAC/AIFF/OGG) and normalize channels to stereo.
2. Mid/Side transform: mid = (L + R) / 2, side = (L - R) / 2.
3. Framing: Overlapping blocks of length 2N with hop N (50% overlap), zero-padded on both ends.
4. Sine window: Apply MDCT analysis window to each block.
5. MDCT: Multiply with the cached cosine basis matrix.
6. Quantize: Scale by 32768 and round to int32.
7. Compression: Zero the last M coefficients per block (where 0 ≤ M ≤ N).
8. Bit-packing: Store coefficients with sign-magnitude encoding into a .bin file.
9. Decompression: Inverse steps via IMDCT + overlap-add + Mid/Side to Left/Right.

Project Structure

• GUI.py: Main GUI (file selection, parameters, log, playback).
• compression.py: Full compression pipeline + helpers (blocking, window, MDCT, bit-writer).
• decompression.py: Full decompression pipeline + helpers (bit-reader, IMDCT, overlap-add).
• audio_stream.py: Low-level audio streaming for playback.
• playback.py: Playback track UI logic (play/pause/seek + time display).
• logging_configuration.py: Color console + file logging setup.
• Test/test_compression.py: Compression unit + integration tests.
• Test/test_decompression.py: Decompression unit + integration tests.
• Analysis/performance_analysis.py: Benchmarking and plotting.
• Data/: Input audio assets.
• Compressed/, Decompressed/, Channels/: Outputs created during runs/tests.

Compression & Decompression Implementation

Compression (compression.py):

• load_stereo_and_compute_mid_side() loads audio, normalizes channels, computes Mid/Side.
• divide_signal_into_blocks() pads and creates overlapping blocks of size 2N.
• generate_sine_window() and apply_sine_window_to_blocks() apply the analysis window.
• generate_mdct_cosine_matrix() and apply_mdct_transform_to_blocks() compute MDCT and quantize.
• save_compressed_mdct() writes a custom header and bit-packed coefficients.

Decompression (decompression.py):

• load_compressed_mdct() reads header + bitstream and reconstructs coefficient matrices.
• apply_imdct_transform_to_blocks() performs IMDCT and synthesis windowing.
• reconstruct_signal_from_blocks() overlap-adds and trims padding.
• sound_decompression() converts Mid/Side back to Left/Right and clips to [-1, 1].

GUI Features

• Audio file and compressed file browsing.
• Adjustable Block size (N) and Compression factor (M).
• Compress, Decompress, and Combined workflow.
• Activity log with timestamped messages.
• Playback section:
  • Original and Decompressed tracks.
  • Play/Pause, click/drag seek, and time display.
• UI remains responsive during processing; non-playback controls are disabled while processing.

Logging (Console + GUI)

• logging_configuration.py creates two loggers: compression and decompression.
• Logs go to:
  • Console (colorized levels via colorlog).
  • Files in Logs/: compression_log.txt, decompression_log.txt.
• GUI has an independent activity log for user-visible actions and errors.

Tests

Located in Test/:

• Compression tests cover block framing, sine windowing, MDCT output, coefficient zeroing, caching immutability, and end-to-end compression.
• Decompression tests cover header decoding, IMDCT formula correctness, overlap-add reconstruction, error handling, and round-trip checks.

Performance Analysis

Analysis/performance_analysis.py:

• Runs compression + decompression across multiple N/M combinations.
• Writes sound_performance_metrics.csv.
• Generates timing plots:
  • sound_compression_times.png
  • sound_decompression_times.png

Usage Notes

• Required packages: numpy, soundfile, customtkinter (optional), sounddevice (playback), colorlog, matplotlib (analysis).
• Run GUI: GUI.py or Sound compressor.exe.