Open-Source and Research-Focused TTS Models and Commercial Solutions
The Awesome-llms-labs/Awesome-llm-tts repository curates open-source and research-focused text-to-speech (TTS) models, including Fish Speech and Kyutai TTS, alongside commercial solutions like WellSaid Labs and Descript Overdub.
Open-Source and Research-Focused TTS Models
The Awesome-llms-labs/Awesome-llm-tts repository curates open-source and research-focused text-to-speech (TTS) models, offering a comprehensive resource for developers and researchers. Among the highlighted projects is Fish Speech, an open-weight multilingual TTS series from Fish Audio, which permits free use for research and non-commercial purposes, with commercial applications requiring a separate license. Another notable entry is Kyutai TTS, a streaming TTS line featuring a 1.6B-parameter model and a 0.75B public-data variant, built on the delayed-streams multistream architecture. For training data, Emilia provides a 101k-hour multilingual in-the-wild speech dataset, available for non-commercial research and educational use only. Multilingual LibriSpeech (MLS) offers a ~50k-hour ASR-derived corpus in eight languages, while Common Voice from Mozilla includes a crowdsourced multilingual corpus with over 100 languages, though it requires heavy filtering for TTS due to variable audio quality.
Commercial and Enterprise TTS Solutions
Commercial and enterprise text-to-speech (TTS) solutions are increasingly being adopted for professional voiceover production and AI-driven content creation. WellSaid Labs offers studio-quality AI voice avatars with over 200 pre-built voices and AI Director pronunciation control, accessible via a documented REST API, making it suitable for enterprise-level voiceover production. Descript Overdub provides voice cloning capabilities that are integrated into Descript’s text-based audio and video editing platform, allowing users to generate new speech by typing in their own cloned voice. These tools are tailored for businesses requiring high-quality, customizable voice outputs. Notably, Play.ht / PlayAI was excluded from this list due to its platform shutdown on December 31, 2025, following a Meta acqui-hire, as reported by Bloomberg on July 12, 2025.
Notable Exclusions and Shutdowns
The Awesome-llms-labs/Awesome-llm-tts repository highlights several notable exclusions and shutdowns in the LLM-era text-to-speech (TTS) landscape. Replica Studios ceased operations on June 30, 2025, with its site now offline. LMNT shut down as of October 1, 2026, with its website displaying a shutdown notice. Coqui (company) closed in January 2025, though its TTS repository remains community-maintained. Papercup’s IP was acquired by RWS in June 2025, ending its status as an independent TTS product. rhasspy/piper was archived in October 2025, with the OHF-Voice/piper1-gpl fork now listed as the maintained version. fairseq is archived and read-only, with TTS coverage directed to live toolkits. TensorFlowTTS and FastSpeech2 implementations are marked as stale and superseded by other maintained projects.
Sources and reporting notes
Based on published source material. No independent hands-on testing is claimed.
