Show HN: Vocal Slice cuts audio clips by selecting transcript text
A Show HN post introduces Vocal Slice, a desktop app for Mac and Windows built around one workflow: instead of scrubbing a waveform to find the part of a recording you need, you load the audio, let it transcribe locally with Whisper down to word-level timestamps, then highlight the phrase you want in the text. The waveform view jumps straight to that region, with draggable start and end handles for frame-accurate trimming, and the selection exports as a named clip. The app accepts WAV, MP3, FLAC, M4A, AAC or OGG input of any length. A WAV source is cut byte-perfect at the byte level, so the exported clip is the original audio rather than a re-encode, preserving the source's exact channel count, sample rate and bit depth; other formats decode to 24-bit WAV on export. Any previously exported slice can be reopened and re-trimmed, since the transcription stays loaded and a second pass costs only the adjustment. Export filenames follow a user-defined template such as {source}{index}{slug}. Whisper transcription supports English and multiple other languages, from Spanish and German to Japanese, Arabic and Hindi, with a choice of model sizes trading speed for accuracy. Processing uses WebGPU when the hardware supports it, including the GPU built into every Apple Silicon Mac, and falls back to CPU automatically. The developer emphasizes that transcription and slicing both happen on-device: there is no account to create and audio is never sent anywhere, a design pitched at unreleased episodes, NDA-covered dialogue or protected interviews. The app costs $29 a year for every feature, all updates, and up to three machine activations, preceded by a seven-day free trial that needs neither an account nor a card. Requirements are Windows 10 or 11 64-bit, or macOS 11 Big Sur and later; any WebGPU-capable GPU or CPU fallback; about 200 MB of disk space for the app plus 75 to 500 MB per transcription model; and an internet connection only once, to download a model and activate the license, not for transcribing itself. The Windows build is not yet signed with a code-signing certificate, so Windows SmartScreen shows an unknown-publisher warning that users must click through with More info then Run anyway.
Key facts
- Vocal Slice transcribes audio locally with Whisper down to word-level timestamps, then lets a user cut a clip by highlighting the words in the transcript.
- WAV exports are byte-perfect cuts of the original file rather than re-encodes, preserving source channel count, sample rate and bit depth; other formats decode to 24-bit WAV.
- Processing runs entirely on-device, including on the GPU built into every Apple Silicon Mac via WebGPU, with automatic CPU fallback, and no audio is ever transmitted.
- Pricing is $29 a year for all features and up to three machine activations, after a seven-day free trial requiring no account or card.
- The Windows build is not yet code-signed, so it triggers a SmartScreen unknown-publisher warning on first run.
Why it matters
Cutting a spoken clip usually means scrubbing a waveform by ear to find where a phrase starts and ends. Vocal Slice inverts that: it transcribes the audio first, so the search happens in text, and the waveform selection follows the words rather than the other way around. That is a narrow but real workflow improvement for anyone who spends hours a week hunting for a specific line in long recordings.
Who it affects
The listing targets podcasters, video editors, voiceover artists and content creators, people who work through hours of raw recordings to isolate the handful of moments that matter. The on-device design also speaks to anyone handling audio under confidentiality constraints, such as unreleased episodes, NDA-covered dialogue or an interview promised to stay private, since nothing leaves the machine.
How to use it
The app is $29 a year, covering every feature, all updates and up to three machine activations, with a seven-day free trial that asks for neither an account nor a card. It runs on Windows 10 or 11 64-bit and macOS 11 Big Sur or later, needs a WebGPU-capable GPU (or falls back to CPU), roughly 200 MB of disk space for the app itself plus 75 to 500 MB per transcription model, and requires internet only once, to download a model and activate the license. The Windows installer currently triggers a SmartScreen unknown-publisher warning because it lacks a code-signing certificate.
How solid is it
This is a Show HN listing, essentially product marketing copy, with 7 points and 3 comments at the time of writing and no independent benchmarks, comparisons, or reviews cited in the source. The described mechanics, word-level Whisper transcription, byte-perfect WAV trimming, WebGPU acceleration, are internally consistent and plausible given how Whisper and WebGPU work, but they come entirely from the vendor's own description rather than third-party verification.
Risks and caveats
No developer name, company background, release date, version history or transcription accuracy benchmarks are given in the source, so there is no way to independently verify performance claims or track record. The unsigned Windows build means every new user on that platform hits a SmartScreen warning before they can run it. As with any subscription software, the seven-day trial converts to a paid $29 annual license if the user continues past it.
“Your audio never leaves your device.”
— Vocal Slice product page