All MicroEvals
Write a complete, production-ready Python script using Modal...
Create MicroEval

Write a complete, production-ready Python script using Modal...

Prompt

Write a complete, production-ready Python script using Modal.com and FastAPI to create an intelligent audio remastering service capable of restoring and mastering old, low-bitrate, or heavily compressed music files. Configure a custom Modal container image pre-installed with FFmpeg, PyTorch, AudioSR, SciPy, Librosa, SoundFile, and any required signal processing libraries, and expose a FastAPI endpoint that accepts audio file uploads in WAV, M4A (both AAC and ALAC), FLAC, or MP3 formats. Decode and normalize all incoming audio into high-precision 32-bit floating-point arrays, and allocate an appropriate NVIDIA GPU (such as an A10G or L40S) to ensure fast inference during neural processing. The Modal image should use an Ubuntu base instead of Debian, due to the recent archival of Debian dependency repositories. Implement a dual-band spectral restoration pipeline that rebuilds both extremes of the frequency spectrum without degrading the original musical core. For high-frequency reconstruction, run AudioSR on the GPU using overlapping audio chunks with smooth crossfading to prevent memory errors on full-length tracks, then use a crossover filter to blend only the newly synthesized upper frequencies above the original Nyquist cutoff back into the untouched low and mid frequencies of the source track. For low-frequency reconstruction, engineer a custom mathematical DSP algorithm combining phase-locked sub-harmonic synthesis—isolating fundamental bass frequencies via bandpass filtering and generating clean sub-octave harmonics—with subtle nonlinear harmonic waveshaping and dynamic low-shelf filtering to restore deep sub-bass weight and psychoacoustic punch while maintaining strict phase coherence. Build an adaptive artifact-removal stage that cleans up lossy compression flaws—such as MP3 pre-echo, metallic "birdie" whistle tones, harsh resonant peaks in the upper mids, and broadband tape or static noise—using dynamic resonant suppression, adaptive spectral gating, and transient-preserving smoothing so the audio is clarified without sounding muffled or dull. Follow this with an intelligent, studio-grade mastering chain that analyzes the track's spectral balance, crest factor, and integrated LUFS to dynamically configure corrective tonal EQ, transparent multiband compression, mid-side stereo imaging (keeping frequencies below 120 Hz strictly mono), subtle analog-style harmonic saturation, and an oversampled true-peak brickwall limiter targeting commercial loudness standards without clipping. Finally, configure the FastAPI endpoint to encode the processed audio into a lossless, studio-resolution 24-bit/48kHz WAV file and return it directly as a downloadable response, complete with temporary file cleanup, clear stage-by-stage logging, and robust error handling for corrupted or unsupported inputs. Provide the entire self-contained Modal script ready for deployment, ensuring every DSP and neural processing step is fully implemented rather than stubbed out.

Drag to resize
Drag to resize
Drag to resize
Drag to resize