Coqui

Coqui (coqui.ai) developed text-to-speech technology combining an open-source deep learning toolkit with hosted commercial products. Its flagship open-source project, 🐸TTS, is a Python library for advanced text-to-speech generation distributed under the coqui-ai GitHub organization and on PyPI. The library provides pretrained models covering more than 1,100 languages, tooling for training and fine-tuning models in any language, and utilities for TTS dataset analysis and curation. It implements a broad range of published architectures, including spectrogram models (Tacotron, Tacotron2, Glow-TTS, Speedy-Speech, Align-TTS, FastPitch, FastSpeech/FastSpeech2, SC-GlowTTS, Capacitron, OverFlow, Neural HMM TTS, Delightful TTS), end-to-end models (ⓍTTS, VITS, YourTTS, plus support for the third-party Tortoise and Bark models), speaker encoders, multiple neural vocoders (MelGAN, Multiband-MelGAN, ParallelWaveGAN, WaveRNN, WaveGrad, HiFiGAN, UnivNet), and voice conversion via FreeVC. Distribution channels included PyPI, Docker images published to GitHub Container Registry, and model hosting on Hugging Face under the "coqui" organization.

Coqui's proprietary model, ⓍTTS, was built on the Tortoise architecture and enabled cross-language voice cloning from a roughly 3-second audio clip, multilingual speech generation at a 24 kHz sampling rate, streaming inference with under 200 ms latency, and fine-tuning. ⓍTTS v2 expanded coverage to 16 languages (English, Spanish, French, German, Italian, Portuguese, Polish, Turkish, Russian, Dutch, Czech, Arabic, Chinese, Japanese, Hungarian and Korean), added support for cloning from single or multiple reference files, and was released under the Coqui Public Model License. The same model powered the company's commercial offerings, Coqui Studio and the Coqui API, which added features such as prompt-to-voice generation, voice fusion, an advanced speech editor for pitch, loudness and emotion control, and team/enterprise controls.

The SaaS product was discontinued in late 2023, with a maintainer stating the team wanted to focus on open-source and model development while selling commercial licences for the XTTS models. In January 2024 the company itself was reported as shutting down via a notice on coqui.ai.

Business model

Coqui paired freely available open-source software with paid commercial products: a hosted studio and API for voice creation and cloning, and commercial licences for the proprietary XTTS models. The open-source 🐸TTS library and pretrained models served as the technical foundation and distribution channel for the paid offerings.

Credit-based subscription tiers for the hosted Coqui Studio product were described as a free trial with 1,800 credits, a Starter Standard tier at $20 for 14,400 credits with API access, a Pro tier (waitlisted) adding team collaboration, higher-quality voice clones and multilingual synthesis, and an Enterprise tier adding SSO, RBAC, audit logs, VPC hosting and custom integrations. After the SaaS product was discontinued, the company sold commercial licences for its XTTS models.

Traction

The 🐸TTS GitHub repository recorded about 46,000 stars, 6,100 forks and 4,668 commits. On Hugging Face, the coqui organization listed 12 team members, three Spaces and two public models, one of which recorded roughly 7.94 million downloads and 3,750 likes as of its December 2023 update.

Latest developments

Coqui released ⓍTTS in 2023 with 13 languages, followed by ⓍTTS v2 with 16 languages, fine-tuning recipes and sub-200 ms streaming. The hosted SaaS was shut down in late 2023 as the team refocused on open source and model development while selling XTTS commercial licences, and in January 2024 a shutdown notice for the company was reported on coqui.ai. The open-source repository and documentation (TTS 0.22.0) remain publicly available, and third-party services continue to offer XTTS V2-based text-to-speech.