Your voice.
Your machine.
Your rules.
Transcribe audio and video, edit against the waveform, and generate speech with models that run on your Mac. Models download when you choose them; projects remain in local app storage.
macOS 14 Sonoma or later · Apple Silicon recommended · Public terms pending
At a glance
What is VoxForge?
VoxForge is a native Mac workspace for local speech-to-text, waveform-synchronised transcript editing and text-to-speech. It downloads models the user chooses, then transcribes and synthesises speech locally in the documented workflow.
- Platform
- macOS 14 Sonoma or later; Apple silicon recommended
- Best for
- Private transcription, subtitle work, transcript editing and local narration
- Processing
- Transcription and speech synthesis use local models
- Availability
- Release validation; public build and commercial terms pending
Product facts reviewed against the current application repository on . See the verification method and corrections policy.
One native Mac app for transcription and speech — fully offline.
VoxForge is a native macOS application that unifies speech-to-text (STT), text-to-speech (TTS), and media transcription into a single tool that runs entirely on your Mac. Transcription is powered by WhisperKit — Apple-Silicon-optimized Whisper that covers 99 languages, produces word-level timestamps, and labels who is speaking with SpeakerKit diarization. For voice output, VoxForge ships more than a dozen local TTS engines — including Qwen3-TTS, Kokoro-82M, and Piper — so you can read any text or transcript aloud in natural, multilingual voices.
VoxForge brings transcription, waveform-synchronised editing, local speech generation, a searchable project library, batch processing, subtitle export, and a multi-track Story Editor into one Mac app. Speech models run locally, while selected model downloads and explicit media fetching use the network. The current repository has no VoxForge account or analytics. VoxForge is built by HighRoad Software in Belgium and is in release validation.
Transcribe, edit, and speak from the same Mac app.
The transcript editor, voice studio, and narration workspace below are from the current VoxForge build.
Everything speech, on one machine
Production-grade transcription and natural voice synthesis, designed to feel native to macOS and to work without a network connection.
Transcribe anything, in 99 languages
Drag in an audio or video file — or transcribe live from your microphone — and VoxForge creates a transcript with word-level timestamps using a selected local WhisperKit model. Accuracy depends on the model, language, recording quality, speakers and domain; no universal word-error rate is claimed.
- Auto-detect language or pick from 99, with optional translation to English
- Speaker diarization labels every voice (SpeakerKit / Pyannote v4)
- Choose model size from tiny to large-v3 to match your Mac's RAM
- Live streaming transcription and batch queues for whole folders
Twelve local voice engines, your pick
Type or paste text and VoxForge speaks it in natural voices — no cloud TTS, no per-character billing. Browse engines by quality, size, and language, preview before downloading, and stream playback so you hear the first words almost instantly.
- Qwen3-TTS (10 languages, optional emotion / style control)
- Kokoro-82M — compact local voice model; verify current model size and language support in the build
- Piper — hundreds of voices across 20+ languages
- Read-aloud any transcript for hands-free proofreading
Waveform-synced editing, Descript-style
Click the waveform to jump the transcript; click the transcript to scrub the audio. Rename speakers, fix words inline, then export to whatever your workflow needs. A multi-track Story Editor lets you stitch narration and clips into a finished piece.
- Export TXT, SRT, VTT, JSON, CSV, DOCX and PDF
- Transcript summaries and translation, generated on-device
- Subtitle burn-in and multi-track timeline
- Project library with full-text search across everything
From recording to result in three steps
No setup gymnastics. Open the app, point it at your audio, and everything runs locally.
Import or record
Drag in an audio or video file, record from your mic, or — in VoxForge Pro — paste a video URL. Models download once on first use and stay cached locally.
Transcribe or synthesize
WhisperKit transcribes with speaker labels and timestamps, or pick a TTS engine to turn text into speech. Everything runs on the Neural Engine and GPU.
Edit & export
Refine the transcript against the waveform, summarize or translate it, then export subtitles, documents, or audio — all without an internet connection.
Built for people who can't send audio to the cloud
Podcasters & creators
Transcribe interviews, label speakers, export YouTube-ready SRT subtitles, and voice B-roll narration — one app instead of four.
Researchers
Accurate multilingual transcription with reliable diarization, batch processing, and structured exports for qualitative analysis.
Legal & medical
Run speech processing locally. You remain responsible for recording consent, professional confidentiality, secure storage, and any model or media downloads you initiate.
Accessibility users
High-quality local voices read documents aloud, with full VoiceOver support and keyboard-first navigation throughout.
Students & journalists
Turn lectures and interviews into searchable, editable text in minutes, then read it back to catch what you missed.
Multilingual teams
99 transcription languages plus natural voices across 20+ output languages — with translation to English built in.
Speech off the grid
Speech inference and project storage are local. There is no VoxForge account or analytics SDK in the current repository. Network access is used for models you choose to download, purchase or licence validation where applicable, and media fetched only from a URL you provide in the direct configuration.
STT + TTS in one place — finally
VoxForge brings transcription and speech generation into one local-first Mac workspace. Competitor capabilities and commercial terms can change, so verify them directly before choosing.
| Capability | VoxForge | MacWhisper | Speechify | Descript |
|---|---|---|---|---|
| Speech-to-text | ✓ | ✓ | — | ✓ |
| Text-to-speech | ✓ | — | ✓ | ✓ |
| Speaker diarization | ✓ | Partial | — | ✓ |
| Waveform-synced editing | ✓ | — | — | ✓ |
| Local speech inference | ✓ | ✓ | Check vendor | Check vendor |
| Works offline | ✓ | ✓ | — | — |
| Commercial terms | Pending release | Check vendor | Check vendor | Check vendor |
| No account required | ✓ | ✓ | — | — |
Comparison reviewed 14 July 2026. Verify current capabilities and terms on each vendor's official product page. Product names are trademarks of their respective owners.
Release channels in validation
The repository contains Store and direct-build configurations. Public prices and purchase links will appear only after those endpoints are live.
- Speech-to-text in 99 languages
- Speaker diarization & timestamps
- 12+ local text-to-speech engines
- Waveform-synced transcript editing
- Import local audio & video files
- Subtitle, document & audio export
- Project library with full-text search
- Everything in the sandboxed Store configuration
- Video-URL import from 1,500+ sites
- Playlist download & batch transcribe
- Extended formats: WebM, MKV, AVI, OGG
- Priority feature requests
Speech inference is local in both configurations. The direct configuration can fetch media you explicitly request; model delivery and entitlement checks also use the network.
Questions, answered
Speech-to-text and text-to-speech use local models on the Mac. Network access is used for models you choose to download, purchase or licence validation when applicable, support links, and direct media fetching only when you provide a URL.
99 languages via WhisperKit — the same coverage as OpenAI's Whisper — including English, French, German, Spanish, Italian, Portuguese, Arabic, Chinese, Japanese, Korean, Russian and Hindi. Language can be auto-detected or set manually, and speech can be translated to English during transcription.
Yes. VoxForge performs speaker diarization with SpeakerKit (based on Pyannote v4), automatically separating and labelling who said what. You can rename speakers, and the labels stay synced with the waveform and word-level timestamps so you can jump to any moment in the audio.
VoxForge combines transcription, local speech generation, speaker labels, and waveform-synchronised editing in one Mac workspace. Competitor features, data flows, and prices change; verify them on their current official pages.
The repository documents multiple local speech engines, including Qwen3-TTS, Kokoro-82M, Piper and Apple system voices. Engine availability, model size, languages and style controls depend on the current build and chosen download. Preview a voice on representative text before selecting it for a project.
Yes. VoxForge transcribes audio and video with timestamps and can export subtitle and document formats. The sandboxed configuration imports local files; the direct configuration can also fetch media from a URL you supply, subject to release availability and third-party terms.
Final commercial terms are not yet live. Any price, renewal model, included features, and refund route will be shown at the verified checkout before purchase.
The repository contains a sandboxed Store configuration and a direct configuration that can add explicit URL fetching and extended formats. Final channel availability and feature boundaries will be confirmed at release. Speech inference remains local in both configurations.
macOS 14 Sonoma or later. Apple Silicon (M-series) is strongly recommended because transcription and synthesis run on the Neural Engine and GPU. Models download once and cache locally — the recommended Whisper Turbo model and Kokoro voice are a small download, and you can choose lighter models on 8 GB Macs.
