AudioEnhancerMAX 3.7.0

Release update: 28 August 2026 Mac App Store candidate build: 3.7.5 Target: Apple Silicon, macOS 13 or later

Release headline

Local-first audio production, now reviewable from diagnosis to export.

AudioEnhancerMAX 3.7 is a substantial workflow release. It moves the app from a collection of processing tools toward a reviewable production environment: the app diagnoses measurable audio risks, proposes deterministic processing, lets the user approve transcript-linked edits, records what happened, and exports through explicit delivery profiles.

Build 3.7.5 stability update

Build 3.7.5 corrects the macOS filter crash reported in the previous package. The crash dump showed an invalid code-signing page while librosa's Numba/llvmlite JIT path was active. The shipping runtime now uses packaged NumPy/SciPy DSP helpers for resampling, spectral analysis, tempo/MFCC support, silence analysis, and frequency restoration. It excludes librosa, Numba, llvmlite, Coqui TTS, and unavailable neural engines instead of requesting JIT or unsigned-executable-memory entitlements. Edge workers now advertise only filters they implement, failed remote chunks fall back locally, and overlap-add preserves duration.

Eight foundations delivered

1. Smart Enhance 2

• Measures level, estimated signal-to-noise ratio, clipping, silence, DC offset, spectral balance, bandwidth and possible 50/60 Hz hum locally.
• Offers Conservative, Balanced and Maximum processing profiles.
• Selects DSP parameters from deterministic measurements; Ollama/Gemma does not control the processing chain.
• Discloses the engine or fallback actually available in the installed distribution.

Smart Enhance is assistive analysis, not a guarantee that damaged source audio can be perfectly reconstructed. Results should be reviewed with A/B listening.

2. Precision Speech Editor

• Links Faster-Whisper word timestamps to an Edit Decision List (EDL).
• Generates review candidates for filler words, repetitions, pauses, breaths and mouth sounds.
• Supports manual word/range selections for cut or duration-preserving mute decisions.
• Defaults candidates to Keep; only user-approved changes are rendered.
• Provides original/proposed preview, persisted bounded undo/redo, and JSON/CSV EDL export.
• Preserves the uploaded source and renders edits to a new output.

3. Native macOS experience

• Runs inside an AudioEnhancerMAX-owned AppKit window powered by WKWebView.
• Does not open the default web browser.
• Uses the AudioEnhancerMAX application menu and product-specific About window instead of Python identity.
• Uses native Open and Save panels for sandbox-safe file access.
• Starts a private service on 127.0.0.1 and terminates it when the app quits or its main window closes.

4. Explicit delivery profiles

Five fixed starting profiles reduce export guesswork:

• Podcast: lossless WAV, normalized to -16 LUFS.
• YouTube/video: WAV, normalized to -14 LUFS.
• Audiobook: lossless WAV, normalized to -18 LUFS.
• Lossless archive: FLAC with source loudness preserved.
• Source-quality master: WAV with source loudness preserved.

WAV/MP3/FLAC, normalization and target-loudness overrides are validated and recorded in the processing report.

5. Processing and governance reports

• Creates a local JSON report for processed and speech-edited exports.
• Records input/output summaries, selected options, resolved output profile, actual pipeline engines, warnings and privacy state.
• Presents a readable, printable report in the interface while retaining the canonical JSON download.
• Records whether optional external or trusted-LAN processing was used.

Quality indicators are diagnostic summaries, not laboratory certification or an objective promise of perceptual improvement.

6. Enforceable runtime boundaries

• Independent settings control external services, local Ollama editorial assistance and trusted-LAN processing.
• Trusted-LAN discovery and processing are disabled by default.
• Core DSP and packaged Faster-Whisper transcription remain local.
• Microsoft Edge Neural TTS is optional and sends only user-entered synthesis text after an explicit request.
• Local Ollama is optional and restricted to editorial prose/expressive rewrite assistance.
• The capability registry reports packaged engines, fallbacks and unavailable features instead of presenting them as universally available.

7. Delivery Compliance Lab

• Measures integrated loudness using the BS.1770 method, plus a 4× oversampled true-peak estimate, sample clipping, sample rate, channels, and duration.
• Provides transparent targets for podcast, online video, audiobook, EBU R128 broadcast, source-master, and archive workflows.
• Returns rule-level pass, warning, or failure evidence and preserves the result locally.
• Clearly labels the result as workflow guidance rather than regulatory or third-party certification.

8. Evidence Chain, Project Capsule, and Review Deck

• Records SHA-256 provenance for the original and processed artifacts.
• Exports a portable ZIP containing the processed output and available processing report, compliance result, transcript, EDL, and blind A/B decision.
• Excludes original audio from the capsule by default; inclusion requires an explicit user choice.
• Adds a focused Review Deck for delivery checks, randomized blind A/B listening, evidence inspection, and capsule export.
• Stores blind A/B results as subjective user-review evidence, never as an objective quality score.

Packaging changes

• Marketing version is 3.7.0; the current App Store candidate is build 3.7.5.
• The App Store package targets native arm64 and macOS 13 or later.
• Faster-Whisper Medium and Silero VAD assets are bundled for local transcription without runtime model download.
• PyAV provides embedded FFmpeg libraries for media decoding and encoding in the Store distribution.
• The package contains the privacy manifest, sandbox entitlements, third-party notices and component licenses.
• Build 3.7.5 embeds the explicit Mac App Store Connect provisioning profile required for TestFlight and retains a single regular privacy manifest under Contents/Resources.
• Demucs source separation and Kokoro voice cloning are not exposed in the Mac App Store build when their required runtime assets are unavailable.
• The separate SourceForge DMG is signed with Developer ID Application, uses hardened runtime and secure timestamps, and was accepted by Apple notarization and stapled on 28 August 2026.
• Gatekeeper accepts the stapled DMG as Notarized Developer ID; notarization submission ID: de84af1d-912e-4270-9919-45659384d2fe.
• The SourceForge Android worker asset is version 1.2.0 (versionCode 3) and remains explicitly labelled as a debug companion build.

Compatibility and boundaries

• Apple Silicon only for the current macOS package.
• Core source execution remains available on supported Python environments, but packaged capability availability can differ.
• Transcript accuracy depends on language, recording quality, overlap and source conditions; high-stakes transcripts require human review.
• Deterministic bandwidth extension does not recreate original missing information.
• Approximate MFCC speaker-change fallback must not be marketed as production-grade neural diarization.
• App Store availability must not be claimed until Apple approves the submitted build.

Validation evidence

The final 3.7.5 source is covered by 51 automated tests, including JIT-free DSP, edge distribution, packaging, build metadata, health/preflight, delivery-threshold, silence, clipping, SHA-256 manifest, source opt-in, ZIP path-safety, and 16 kHz AutoEQ signal-preservation checks. The hardened candidate and final direct-download app both completed the supplied 8.41975-second WAV through a 12-step filter chain, produced a 48 kHz output of identical duration, and remained alive. The final 3.7.5 PKG passed installer signature, deep payload signature, embedded-profile, bundle/version, privacy-manifest, 332-file arm64, macOS 13 deployment-target, absolute-dependency, and entitlement checks. The direct-download DMG additionally passed Apple notarization, stapling, Gatekeeper assessment, deep signature verification, launch, health, and final filter processing. Transporter delivered the 3.7.5 App Store package without warning on 28 August 2026 at 12:31 CEST and reports it ready for internal testing; TestFlight installation, physical-Mac recording, and Apple review remain separate gates.

See:

• releases/marketing/CLAIMS_EVIDENCE_MATRIX_v3.7.0.md
• releases/marketing/APP_STORE_METADATA_v3.7.0.md
• releases/marketing/LAUNCH_CAMPAIGN_v3.7.0.md
• CHANGELOG.md