{"name":"ema-lightning","repo":"canberk7/ema-lightning","url":"https://midorreal.com/project/canberk7-ema-lightning","repository":"https://github.com/canberk7/ema-lightning","description":"Tiny, fast and accurate Turkish text-to-speech. 8.6M parameters, 0.92% WER on Freya-TR-Eval, first audio in ~4 ms and 1,300× real time on one GPU. Streams audio, batches many callers on one GPU, and runs offline on a GPU or CPU.","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/canberk7/ema-lightning","checked":"9 Oct 2026"},"language":{"value":"Python","source":"https://github.com/canberk7/ema-lightning","checked":"9 Oct 2026"},"last_release":{"value":"v1.0.4, 7 Oct 2026","source":"https://github.com/canberk7/ema-lightning/releases","checked":"9 Oct 2026"},"releases":{"value":"4","source":"https://github.com/canberk7/ema-lightning/releases","checked":"9 Oct 2026"},"contributors":{"value":"2","source":"https://github.com/canberk7/ema-lightning/graphs/contributors","checked":"9 Oct 2026"},"archived":{"value":"no","source":"https://github.com/canberk7/ema-lightning","checked":"9 Oct 2026"}},"stars":152.0,"writeup":{"what_it_is":"EMA Lightning is a Turkish text-to-speech system with 8.6M parameters that runs offline on GPU or CPU. It generates speech from text with 0.92% word error rate and produces the first audio in about 4 ms.","audience":"Developers building Turkish voice applications offline","claims":[{"kind":"specific","claim":"Achieves 0.92% word error rate on Freya-TR-Eval benchmark","excerpt":"0.92% word error rate, the lowest of every system measured, including ElevenLabs v4, Gemini 3.8 and the 2.38B-parameter Trendyol-TTS.","status":"not_checked"},{"kind":"specific","claim":"First audio ready in approximately 4 milliseconds","excerpt":"the first audio is ready in about 4 ms","status":"not_checked"},{"kind":"specific","claim":"Runs 440× faster than real time on single request with RTX 4090","excerpt":"One request runs 440× faster than real time, and batched work 1,316× (RTX 4090).","status":"not_checked"},{"kind":"specific","claim":"Model size is 8.6M parameters, approximately 34 MB","excerpt":"8.6M parameters, about 34 MB","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-10-09"},"why_now":null,"owner_supplied":null,"score":{"verdict":"mid","hype":38,"reality":33,"gap":6,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-10-09T05:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-10-07T22:37:32+00:00","text":"Latest release v1.0.4","source":"https://github.com/canberk7/ema-lightning/releases"},{"date":"2026-10-05T21:41:34+00:00","text":"First release","source":"https://github.com/canberk7/ema-lightning/releases"},{"date":"2026-10-05T20:19:47+00:00","text":"Repository created","source":"https://github.com/canberk7/ema-lightning"}],"cite":"ema-lightning is a GitHub project at https://github.com/canberk7/ema-lightning written mainly in Python, described there as \"Tiny, fast and accurate Turkish text-to-speech. 8.6M parameters, 0.92% WER on Freya-TR-Eval, first audio in ~4 ms and 1,300× real time on one GPU. Streams audio, batches many callers on one GPU, and runs offline on a GPU or CPU\".","attribution":"Data from Mid or Real, https://midorreal.com/project/canberk7-ema-lightning, checked 9 October 2026. Attribution with a link is required."}