Newest papers

Measuring the Verifier's Floor: How Much Apparent Rare-Word Text-to-Speech Error Is ASR Error?

Kirill Borodin, Vasilii Kudryavtsev, Nikita Vasiliev, Grach Mkrtchian · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31470 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Tekst-to-speech (TTS) biasanya dievaluasi dengan transkripsi outputnya dengan sistem pengenalan suara otomatis (ASR), sehingga tingkat kesalahan kata juga menghitung kesalahan pengenal. Kami mengukur kesalahan ini pada rekaman manusia dari kalimat uji. Pada LibriSpeech dan FLEURS, pengenal evaluasi melewatkan 33-67% dari kata-kata yang sangat langka bahkan dalam bahasa manusia, dan semua tiga pengenal melewatkan 22-42%. Memeriksa setiap kata dengan dua keluarga pengenal lebih rendah kesalahan TTS yang jelas pada kata-kata yang setidaknya satu pengenal mendengar dua sampai tiga kali (sangat langka kata-kata LibriSpeech: 10-13% hingga 3-5%), namun kata-kata yang sangat langka tetap menjadi yang paling sulit. Mengabaikan kata-kata pengenal melewatkan dalam bahasa manusia adalah bias:

Machine translation (Meta M2M100); the English original follows.

Abstract

Text-to-speech (TTS) is usually evaluated by transcribing its output with an automatic speech recognition (ASR) system, so the word error rate also counts recognizer errors. We measure these errors on human recordings of the test sentences. On LibriSpeech and FLEURS, the evaluation recognizer misses 33-67% of very rare words even in human speech, and all three recognizers miss 22-42%. Checking every word with two more recognizer families lowers the apparent TTS error on words that at least one recognizer hears two- to three-fold (very rare LibriSpeech words: from 10-13% to 3-5%), yet very rare words remain the hardest. Ignoring words the recognizer misses in human speech is biased: the other recognizers find TTS errors far more often on these words (4-6% vs. 0.2-0.3%) and overturn about half of the errors this mask keeps. As half of the remaining adjudicated errors may be spelling variants, a listening test is needed.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Speech recognition, Computer science, Word error rate, Word (group theory), Spelling, Active listening

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange