Open TTS Leaderboard adds 31M and 35B models and the UTMOS metric
The Open TTS Leaderboard update compares a 31M-parameter open model with a 35B one on English WER and adds the UTMOS metric for speech naturalness.
Source: Open TTS Leaderboard adiciona modelos de 31M e 35B e a métrica UTMOS (x.com). Text prepared with AI from this source.
What happened and what to do
The Open TTS Leaderboard, in an update published on 23/10/2026, now includes two open text-to-speech models of very different sizes: phonon, with 31M parameters, from Gradium, and Drama-3, with 35B, from FishAudio. In the published results, English WER was close, at 1.08 for the small model and 1.19 for the large one. The update also added the UTMOS metric, which assesses the naturalness of English output.
For a company planning to use synthetic voice in customer service, narration or digital products, the case shows that model size and perceived quality do not always move together, but that this comparison has to be run on the company's own content. A practical implementation would be an evaluation harness that runs candidate TTS models on real client texts, measuring WER, UTMOS and latency on the client's own hardware, with private inference so voice data is not exposed. That harness can become a monitoring pipeline that re-evaluates the model in production whenever new versions appear.
How the consultancy can help
Diagnosis of where speech synthesis fits in the client's process and which quality, cost and latency criteria matter. Then implementation of a comparative evaluation pipeline for open TTS models on the client's own samples, with private inference and a monitoring dashboard.
Next step
If your company is evaluating or already uses synthetic voice, send a short description of the case, including audio volume, language and current hosting environment. We will reply with a scoped proposal for diagnosis and implementation.
Let's assess your project
I reply with scope and a proposal.