Skip to main content

VoiceStudio

VoiceStudio is a free, open-source voice app that runs on your own computer. Reelcraft can use it for:

  • Voice-over in Generate Speech, with any speech engine you install in VoiceStudio.
  • Your own voices: make a voice from a recording in your channel's Library, or use the voices you cloned or designed in VoiceStudio.
  • Word timings for captions, with no separate step.
  • Transcription in Analyze Media and for quality control's Include transcript.

All of it is free: nothing is billed per request. With VoiceStudio on the same computer, no text or audio leaves it. With VoiceStudio on another computer, the text and audio go to that computer.

When VoiceStudio isn't reachable, quality control's Include transcript uses Deepgram instead if you set a Deepgram key, which is paid and sends the audio to Deepgram. See Transcription.

VoiceStudio is a separate app. Reelcraft doesn't install, start or update it: it only talks to it while it is running.

Set it up​

  1. Install VoiceStudio from its download page, or with its install command (see VoiceStudio's Get started). It runs on macOS, Windows and Linux, and in Docker.
  2. Open VoiceStudio. The first time, it offers to download its default speech engine. Let it.
  3. In Reelcraft, open Settings → VoiceStudio. The status at the top says Connected to VoiceStudio, with its version and what it runs on (cuda for an NVIDIA graphics card, mps for a Mac's GPU, cpu otherwise).

If it says VoiceStudio isn't reachable, check that VoiceStudio is open, then see Where Reelcraft looks for VoiceStudio.

Where Reelcraft looks for VoiceStudio​

VoiceStudio listens on port 3900. Reelcraft looks for it at:

Reelcraft runsAddress
Directly on your computerhttp://127.0.0.1:3900
In Docker (the Reelcraft app)http://host.docker.internal:3900

To use another address, for example VoiceStudio on another computer with a graphics card, enter it under VoiceStudio address in Settings → VoiceStudio, select Test connection, then Save. Reset goes back to the default. The address can also be set with the VOICESTUDIO_URL environment variable; one saved in Settings takes priority.

VoiceStudio's desktop app only listens on the computer itself (127.0.0.1), and needs no key for requests from there.

  • Docker Desktop (Mac and Windows) passes host.docker.internal through to your computer, so the default address is expected to work with VoiceStudio's default settings. If the status says it isn't reachable, use one of the Linux setups below.

  • Docker on Linux (Docker Engine, without Docker Desktop) can't reach a program that only listens on 127.0.0.1. Use one of these:

    • VoiceStudio on all interfaces, with a key. Start VoiceStudio's backend with OMNIVOICE_BIND_HOST=0.0.0.0 and OMNIVOICE_API_KEY set to a long random key (see VoiceStudio's API authentication). Start the Reelcraft container with --add-host=host.docker.internal:host-gateway, keep the default address, and enter the same key under API key (optional). Allow port 3900 from Docker's network in your firewall, and not from anywhere else.
    • VoiceStudio in Docker too. Put both containers on one Docker network, save http://<VoiceStudio's container name>:3900 as the address, and start VoiceStudio with OMNIVOICE_API_KEY and enter the same key here.

    Don't start the Reelcraft container with --network host for this: the database, file storage and job runner inside it would then use your computer's own ports, and could clash with other programs or be reachable from your network.

Don't use VoiceStudio's Network sharing (the Local / Network switch) for Reelcraft: it listens on another port and asks for a PIN that changes every time, which Reelcraft can't send.

API key​

VoiceStudio accepts requests from the same computer without a key. If you started VoiceStudio with an API key (OMNIVOICE_API_KEY, see VoiceStudio's API authentication), Reelcraft needs the same key when it reaches VoiceStudio from Docker or from another computer. Paste it under API key (optional) in Settings → VoiceStudio, or set VOICESTUDIO_API_KEY. It is stored encrypted, like the AI provider keys. Without it, the status says VoiceStudio refused the request.

For a VoiceStudio on another computer, use https:// (for example through VoiceStudio's Tailscale support or a reverse proxy) or an encrypted tunnel such as Tailscale or WireGuard. Over plain http:// the key, the text and the audio travel unencrypted, and anyone on the network path can read and reuse the key. The connection test only sends the key to the address that is saved, so save an address before testing it with a key.

Make speech with it​

  1. In a Generate Speech stage (or the Speech and audio default), choose VoiceStudio as the provider.
  2. Choose a Model. Each speech engine installed in VoiceStudio is a model here, such as VoiceStudio (OmniVoice), KittenTTS or IndexTTS 2.5. To get more, install them in VoiceStudio's Model Catalogue, then select Check again.
  3. Choose a Voice: one you saved in VoiceStudio, or one of the engine's built-in voices. Until you choose, the engine's default voice speaks. Engines that can't clone (such as KittenTTS) only show their built-in voices.
  4. Adjust the settings, and select Speak it to hear a sample. Samples are free.

A blueprint keeps the engine you chose, even if you switch the active engine inside VoiceStudio.

Settings:

  • Language: the language code of the text, such as en, es or hi. Leave it empty and the engine works it out.
  • Speed: from 0.5 (slower) to 2 (faster).
  • Delivery: how to say it, in words, such as "calm and warm". Only engines that follow written instructions show it.
  • Word timings: see below.
  • Under Advanced: Seed. The same seed, text and settings make the same take again.

Formats: WAV (default), MP3, FLAC and Opus. VoiceStudio takes about 4,000 characters at a time; a longer script is split at sentence ends and joined, in WAV or MP3.

The first speech from an engine can take minutes: VoiceStudio loads the engine's model, and downloads it the first time. After that it is quick.

Your own voices​

To make a voice from a recording:

  1. Upload the recording to the channel's Library (see Assets). Use a clean recording of one person speaking, about 5 to 30 seconds long. An audio or a video file works.
  2. On its card, select Make a VoiceStudio voice.
  3. Give the voice a name. Optionally type what the recording says and its language; if you leave them empty, VoiceStudio works them out.
  4. Select Make voice.

The voice is saved in VoiceStudio, so it also appears in VoiceStudio's own app. To use it, choose a VoiceStudio engine that can clone voices (such as VoiceStudio (OmniVoice)) in a Generate Speech stage, then pick it from the voice list. Engines that can't clone, such as KittenTTS, only offer their built-in voices.

The file can be up to 50 MB. A short recording is better anyway.

Voices you clone or design in VoiceStudio's own app appear in the list too.

caution

Only clone a voice you have the right to use: your own, or someone's who agreed to it.

Word timings​

Turn on Word timings in a VoiceStudio Generate Speech stage and add a Memory write with the path timing. After making the speech, VoiceStudio transcribes it and Reelcraft saves when each word and sentence is spoken, for captions. It is the same as for ElevenLabs: see Word timings. It is free, and adds a few seconds.

Transcription​

In Analyze Media, choose VoiceStudio as the provider and VoiceStudio transcription as the model to transcribe a voice-over, or any audio or video, with word timings, for free. VoiceStudio uses the transcription engine that is active in it.

Quality control's Include transcript also uses VoiceStudio, when the judge model can't listen to audio itself and VoiceStudio is running. It is used before Deepgram.

When VoiceStudio is closed​

  • The stage editor shows VoiceStudio isn't reachable, with Check again and Open VoiceStudio. A stage keeps the engine you chose.
  • You can still save the blueprint: it shows a warning, not an error.
  • A run goes as far as the first stage that uses VoiceStudio and fails there, saying that VoiceStudio isn't reachable. Open VoiceStudio and retry the stage.

Cost​

Nothing. VoiceStudio runs on your computer. If you want your budget to count something anyway, enter a price under Advanced → Price per 1,000 characters.