Coqui

239 HackerNews comments (top 25 shown, all developer-skewed) describe Coqui as an open-source TTS/STT toolkit from a startup that has since shut down. Praise centers on quality and accessibility: one highly upvoted user raves about the 109 VCTK voices ('p236 male and p237 female are my choices... 109 quality voices still blows my mind'), easy install on WSL with CUDA, and notes that good TTS is now commoditized ('you had to pay for a good TTS just a year ago'). XTTSv2 is ranked 'only slightly behind StyleTTS 2' on the TTS Arena leaderboard and has a low-latency streaming mode users appreciate. But the same commenters report real flaws: XTTSv2 'will hallucinate nonsense words or insert extra syllables in words, pretty frequently'; ElevenLabs is repeatedly called 'so far ahead'; and one user evaluating Coqui as a vendor says 'I went as far as reporting bugs to them, just to have them aggressively ignore me' — though another calls the team 'very friendly and welcoming.' The dominant theme is abandonment plus legal risk: 'Coqui is no longer actively maintained. I'm not sure what the team is up to now'; the code is MPL 2.0 (commercial use OK) but the XTTS model sits under CPML, which 'doesn't allow commercial use,' leaving one user unwilling to build on it ('worried about it being a bit of a dead end'). Developer-experience complaints include Tensorflow/Bazel pain ('turnaround time was about an hour per change... Eventually I gave up'). Users actively recommend alternatives: Piper (lightweight, realtime), MARS5 (newly open-sourced), StyleTTS 2 (better license), Meta Voicebox (research-only). Comments also cover the older Coqui STT model, noting training data mixes free corpora (LibriSpeech, Common Voice) with non-free data (Fisher, Switchboard, ~1700h of NPR/WAMU radio) — enough to be flagged 'toxic candy' under Debian's ML policy. Several threads are off-topic trivia about the Puerto Rican coquí frog the company is named after.

No aggregate ratings were found for this product during the last harvest.