Introduction

Riverside Labs launched Tonality on November 12. Riverside is the remote podcast and video recording platform, and Riverside Labs is its new research lab for audio and video models. Tonality is its first model. CEO Nadav Keyson calls it "the first voice model that edits any word in an existing recording, with full control over how it's said."

Most TTS and voice-cloning tools make you regenerate the whole sentence when you fix one word. That changes the delivery, and the patch often doesn't match the take around it. Tonality instead edits in place. You swap, insert or delete words, or add whole new sentences, inside a real recording. You can also set pitch, tone, duration, pauses, energy and pronunciation on any single word. The speaker's voice, pacing, accent and room acoustics are kept. It's a closed model. The web playground is free with rate limits, and API access goes through sales.

Sources:

(November 12, 2026), the Tonality product page, and the thread's
.

What Shipped

  • Name and maker: Tonality, by Riverside Labs, the research and API arm of Riverside (riverside.com).
  • Modality: speech editing and speech generation. It works on audio files you already have and on typed text, where you pick a reference voice and generate a take on the fly.
  • Word-level controls: pitch, tone, length, pause, energy and pronunciation, set as cues on any individual word. You can also "lock" the parts that are already right.
  • In-line editing: change, insert or remove words, or add new sentences, inside a real recording.
  • Identity preservation: edits and new words keep the speaker's voice, pacing, accent and the acoustics of the room they recorded in.
  • Verification output: besides the edited audio, Tonality returns the exact sample ranges that changed, so you can confirm nothing else moved.
  • Languages: 10+, according to
    . The product page doesn't list them.
  • Open vs closed: closed and hosted. There are no weights, paper or public API docs yet.

Where to Run It Today

  • Playground: open labs.riverside.com/tonality and choose "Try for free." You get every control on your own recordings, rate-limited.
  • API: "Contact sales." The page says it offers volume pricing, with keys, docs and support handed out through sales, and no self-serve signup.
  • Workflow: upload a recording or paste a script, pick a reference voice, edit the transcript, add per-word cues and generate. Then check the returned changed-sample ranges against the original.

What It's Built For

  • Podcast ad insertion: swap a sponsor name or an offer inside an existing episode in the host's own voice and room.
  • Approved takes: change one word in a signed-off VO and keep every other word as recorded.
  • Re-direction: add a lift on one word or a beat before a line without re-recording the read.
  • Audiobooks and narration: fix a mispronounced name or a flat line without re-recording the chapter.
  • Dubbing: lengthen a word or add a pause so a line fits the shot.
  • Agent-made video: an editing agent adds a line of narration while the rest of the track stays untouched.

How It Compares (Vendor Demo)

Riverside's product page runs one script edit through several systems so you can listen side by side:

  • Tonality gets per-word cues (emphasis, insert, replace, pause, rising pitch).
  • Cartesia Sonic 3.6 is driven with SSML-style break and volume tags.
  • ElevenLabs v4 and HeyGen Voice get bracket pauses and capitalized emphasis.
  • Fish Audio Drama 3 gets emphasis tags.
  • Qwen Audio TTS gets plain text.

The point Riverside is making is that the other systems regenerate the whole passage, while Tonality only rewrites the changed words. There's also a dedicated

. These are hand-picked listening demos, not a scored benchmark, so run your own A/B before trusting them.

Caveats

  • No numbers yet: there are no MOS, speaker-similarity or WER figures, and no paper.
  • API is sales-gated: builders can't get a key or pricing from the page today.
  • Consent matters: an editor this precise makes it easy to put new words in a real person's mouth. Use it on your own voice or with the speaker's explicit permission, and keep the original takes.
  • Language coverage is only "10+" in a reply, so test your language in the playground first.

Conclusion

Tonality is a closed, word-level speech editor from Riverside Labs. It patches a word, a pause or an inflection inside a real recording instead of regenerating the whole line, and it keeps the voice and the room sound. If you produce podcasts, VO, audiobooks or dubs, the free playground is worth trying today on a take you already have. Compare it with your current TTS patching against Cartesia or ElevenLabs, and hold off on building against the API until Riverside publishes docs and pricing.

—Anabel ♡