Jonathon's AI Wiki

Fluid Voice and Voice Box — Free Self-Hosted Alternatives to Whisper Flow and ElevenLabs

6 min read

Source: raw/Free_Wispr_Flow_ElevenLabs_CapCut_alternatives_+_more_GitHub_hits.md — YouTube tool-roundup video (youtube.com/watch?v=3Cni6_JubQk, fetched 2026-07-06), sponsored by Zapier. Two hosts (named “Andrew” and “Adam” in dialogue; possibly Andrew Warner and Adam Brakhane per an unconfirmed secondary-search match — the channel name is not stated on-screen in the transcript). Refreshed 2026-07-20 with raw/x-bookmarks-recent-digest-2026-07-21.md (X bookmark, @adiix_official) identifying Voice Box as the open-source repo jamiepine/voicebox.

Two free, self-hosted tools surfaced in a weekly GitHub/AI-tool roundup as direct replacements for paid dictation and voice-cloning subscriptions: Fluid Voice (a Whisper Flow alternative for dictation) and Voice Box (a voice-cloning and text-to-speech tool that also does dictation). Both run locally rather than in the cloud, which the hosts frame as the practical reason to switch — no subscription, no bandwidth round-trip, and nothing you dictate leaves your machine.

Key Takeaways

  • Fluid Voice — free, open-source, self-hosted dictation tool positioned as a direct Whisper Flow replacement. Runs locally (demoed on Mac), corrects transcription as you speak, needs no API keys, and reportedly “handles slang better than expected.” One user quoted in the video (Reddit user “Dilip,” described as a heavy Whisper Flow user — 44,000 words dictated over 6 months, top 1.1% of Whisper Flow users) switched to Fluid Voice and canceled his paid Whisper Flow plan.
  • Local processing is the whole pitch, not a side benefit. Because the full voice model ships inside the app (the hosts note a ~3.5GB download), Fluid Voice needs no network round-trip — the hosts argue this makes it both faster than a cloud-based tool and immune to any “random stuff you say” ever leaving your machine.
  • One host tested it live on-camera (during the recording) and confirmed it worked well enough that he considered canceling his own paid dictation subscription (a different paid tool, “Super Whisper,” not Whisper Flow itself).
  • Whisper Flow’s price point was the explicit contrast — one host called its $15/month fee “stupid” given the free local alternatives now available.
  • Voice Box — a second free tool covering some of the same dictation ground, but its primary function is voice cloning and text-to-speech: turning typed text into speech in a cloned voice. The video’s live demo claims a voice was cloned “in under 30 seconds” from a short sample. Framed explicitly as removing the need to pay for both an ElevenLabs subscription (voice cloning) and a Whisper Flow subscription (dictation) at once.
  • No repo URL, star count, or license was stated in the source for either tool — both were shown via their own websites/GitHub pages on-screen without the hosts reading out identifying details. Verify current details directly before adopting.

Practical framing

The hosts’ argument for switching is entirely cost- and control-based, not capability-based: they don’t claim either tool beats ElevenLabs or Whisper Flow on quality, only that the free/local/self-hosted versions are now “good enough” to justify canceling paid subscriptions. This is the same “local models are closing the gap fast enough to matter” argument the hosts make elsewhere in the same video about coding models (see the companion roundup below) — voice is one of the categories where a full model can now ship bundled inside a desktop app rather than requiring a cloud API.

Update 2026-07-20 — Voice Box is the open-source repo jamiepine/voicebox

A widely-shared X post (@adiix_official; raw/x-bookmarks-recent-digest-2026-07-21.md) supplies the identifying details the original roundup left blank. They match the Voice Box described above — same voice-cloning + TTS + dictation combo, same ElevenLabs/Wispr-Flow-replacement pitch — so this is almost certainly the same tool.^[inferred] Per that post (a single social source; specifics not independently verified):

  • Repo: jamiepine/voicebox (developer Jamie Pine), MIT-licensed, ~41.6k GitHub stars (“in a matter of weeks”).
  • Clones a voice from ~10 seconds of audio; 7 TTS engines to choose from; 23 languages (e.g. Arabic, Japanese, Polish).
  • Local-first via Tauri (Rust, not Electron) — voice and samples never leave the machine.
  • Exposes an MCP server: Claude Code, Cursor, Cline, and any MCP-aware agent can reply in a chosen cloned voice through a single voicebox.speak call — i.e. give your terminal agent a spoken voice.
  • Global dictation hotkey inside any app, positioning it as a Wispr-Flow replacement as well as an ElevenLabs one.

The post’s “THIS DEVELOPER JUST KILLED THE ENTIRE VOICE CLONING INDUSTRY” framing is marketing hyperbole, not a verified market outcome — read it as an enthusiastic launch signal. The checkable facts (MIT license, MCP voicebox.speak, local Tauri runtime, ~10s cloning) are what matter for adoption.

Try It

  1. Search for “Fluid Voice” (dictation) and “Voice Box” (voice cloning/TTS) directly — the source video didn’t give repo URLs, so confirm you have the right project and check its current license, platform support (only Mac was confirmed in the source), and maintenance activity before adopting.
  2. If you’re paying for Whisper Flow ($15/mo per the source) or a similar cloud dictation subscription, trial the local alternative for a week before canceling — confirm accuracy on your own vocabulary/accent, since the source’s only accuracy testimony is two individual users’ impressions.
  3. For voice cloning, test Voice Box against your existing ElevenLabs workflow on a real script before committing — the “under 30 seconds to clone” claim is a single demo, not a benchmark.
  4. Compare against whisper-cli (whisper.cpp), the wiki’s existing free/local/offline Whisper path — Fluid Voice appears to target the same offline-dictation niche with a friendlier out-of-the-box app experience (bundled model, live correction) versus whisper-cli’s CLI-first, six-model-size flexibility.

Open Questions

  • Exact GitHub repos, licenses, and current star counts — Voice Box is now identified as jamiepine/voicebox (MIT, ~41.6k stars per the 2026-07-20 X post above); Fluid Voice’s repo/license/star count remain unstated in any cited source.
  • Platform support beyond Mac (Windows/Linux) for either tool — not addressed.
  • Language support for Voice Box’s cloning/TTS — the roundup demo used English only, but the 2026-07-20 X post claims 23 languages (unverified, single social source).
  • Whether Fluid Voice’s “corrects as you speak” claim holds up across accents/vocabularies beyond the two individual testimonials cited.

Graph View

Backlinks