Court-recording audio, opened and useful.

DepoAudio is a free, open-source desktop app for opening, converting, improving, reviewing, and organizing court-recording audio. The current published release is v1.0.2 for macOS 13.3 or later universal and Windows 10 or 11 x64.

No audio-upload or telemetry path was found in the reviewed v1.0.2 source. Conversion, cleanup, and installed-model analysis run on the device. Selecting Install for an optional model contacts GitHub, downloads a bounded file, and verifies its SHA-256 hash; the recording is not part of that request. The published release was built without an updater endpoint or signing key and has no updater manifest or updater signature assets.

DepoAudio v1.0.2 searches the first 8 KB for an embedded Ogg stream, removes leading application data for supported variants, and converts the audio to a standard output. Keep the source because variants can differ and lossy output re-encodes audio.

FTR/TRM support is experimental in v1.0.2. Preserve every original, try the conversion, and verify the full output. Important decoder and segment-ordering fixes remain in source ahead of the published release; export standard audio with FTR software if the release fails.

The current release cannot decrypt an Eclipse AudioSync file. Export WAV from Eclipse first, then use DepoAudio for conversion, cleanup, channel work, or organization. The website identifier treats the extension as a filename clue rather than proof of the container.

DepoAudio v1.0.2 does not open DCR directly. Use Liberty Court Player to export WAV, then open the exported audio in DepoAudio. Menu names and export options can vary by Liberty version.

DepoAudio opens Broadcast WAV as standard WAV audio, but a converted output may not retain the original broadcast metadata. Keep the source BWF as the record copy when timecode, originator, or other metadata matters.

The published release exports WAV, MP3 at 128/192/320 kbps, FLAC, Opus, and M4A/AAC. It can mix to stereo, keep the original channel layout, or write one file per channel. Channel labels are editable roles, not voice identification.

The app enforces file-size safety limits; v1.0.2 defaults to 2 GB per audio file, and configured conversion limits range from 0.5 to 10 GB. No verified public batch-count capacity has been established, so the website does not claim unlimited files or duration.

RNNoise background-noise reduction, de-clip, and bounded per-channel auto-level are released. The release also includes Silero speech activity, Smart Turn boundaries, active speaker-slot estimates, FlashSR narrowband enhancement, and an optional DNSMOS quality estimate. Silero passed exact-commit macOS and Windows release inference smokes; the other model paths were reviewed for assets, hashes, loaders, and implementation but did not receive separate packaged-model inference smokes. DeepFilterNet remains unsupported because its reserved model files have no production spectral pipeline.

No. v1.0.2 can estimate likely turns and active segmentation-model slots, but it does not count distinct people or identify anyone by voice. Speaker embedding and clustering do not run. Editable labels such as Reporter or Witness are user-assigned workflow labels.

Some v1.0.2 ONNX paths may use CoreML on macOS. Silero VAD and speaker-slot segmentation intentionally run on CPU, and Windows uses the bundled CPU ONNX Runtime. DepoAudio does not claim verified Apple Neural Engine, DirectML, Windows GPU, XDNA, Intel AI Boost, or NPU execution.

Silero VAD, Smart Turn v3, FlashSR, speaker segmentation, unused DeepFilterNet assets, and ONNX Runtime ship with the app. DNSMOS downloads from GitHub only when you select Install. A speaker-embedding asset is also downloadable but unused by any released inference or identification path. Downloads are bounded, written atomically, and SHA-256 verified; audio is not uploaded.

No. DNSMOS estimates speech, background, and overall quality from a fixed 9.01-second mono excerpt. Treat the scores as review guidance, not measured improvement, guaranteed intelligibility, or evidence about the entire recording.

Yes. In the Player, import SRT, VTT, or TXT, paste or type text, follow timed lines during playback, stamp lines to the playhead, and export SRT or TXT. This is local transcript proofing, not automatic transcription or voice identification. Export a copy before moving the audio or clearing app data.

Not in the supported current-release workflow. Merge remains hidden and independently release-gated under app issue #90; retained source code is not downloadable product behavior.

Use the local identifier and try a synthetic or non-confidential reproduction first. If a sample is necessary, remove names, case numbers, voices, paths, and unneeded metadata, share only the smallest useful excerpt when authorized, and assume GitHub issues and attachments are public.