You searched a paid product
Descript does 3 jobs: transcript-based editing, voice cloning, speech generation. Most projects replace one or two of them, so pick the job you actually need.
Open-source alternatives to Descript that run locally, allow commercial use and need no paid API include GPT-SoVITS (voice cloning and speech generation; MIT; runs on CPU) and Dia (voice cloning and speech generation; Apache-2.0; GPU, ~4.4GB+). None of the 9 projects we track covers all 3 Descript jobs, so most people combine two or more.
Resemble AI's family of open TTS models (Turbo, Nano, Multilingual) with zero-shot voice cloning from a reference clip.
Python TTS library with many models (XTTS-v2, VITS, YourTTS, Bark, Tortoise); original repo unmaintained, continued by idiap fork.
1.6B-parameter English text-to-dialogue model from Nari Labs with two-speaker tags, non-verbal sounds and audio-prompt voice cloning.
Flow-matching text-to-speech research model that clones a voice zero-shot from a reference audio clip.
Fish Audio's multilingual TTS and voice cloning stack (Drama 3, S2-Pro) under a source-available community licence.
Few-shot voice cloning and TTS WebUI; 5-second zero-shot or 1-minute fine-tune, five languages, MIT code and weights.
MyShell's instant voice cloning model with tone-colour transfer, style control and cross-lingual cloning; MIT code and weights.
Transcript-based audio and video editor; delete words in the transcript to cut the media. Runs on-device. Non-commercial licence.
Research text-to-speech model using style diffusion and adversarial training, with pretrained LJSpeech and LibriTTS checkpoints.
No project meets all of your requirements. The list below shows what you'd need to relax.
We hid them because something is missing, unknown or conflicting. Here's what.
Yes. GPT-SoVITS, Fish Speech, Chatterbox, F5-TTS and StyleTTS 2 are free to download and run on your own computer without paid API keys. Each covers only part of what Descript does, so check which job you need.
Yes, with Fish Speech (Custom licence), GPT-SoVITS (MIT) and Chatterbox (MIT). Check the hardware notes and licence conditions on each page before you start.
Confirmed: Chatterbox, Dia, GPT-SoVITS and OpenVoice. With conditions: Coqui TTS, Fish Speech and StyleTTS 2. Not for commercial use: F5-TTS and Rescript, usually because the model weights are non-commercial. This is not legal advice; read each licence.
Not as of 2026-10-16. Transcript-based editing, voice cloning and speech generation are covered by different projects.