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

Portuguese

Text-to-speech (TTS) é geralmente avaliado por transcrição de sua saída com um sistema de reconhecimento automático de fala (ASR), de modo que a taxa de erro da palavra também conta erros do reconhecedor. Medimos esses erros em gravações humanas das sentenças de teste. Em LibriSpeech e FLEURS, o reconhecedor de avaliação perde 33-67% de palavras muito raras, mesmo no discurso humano, e todos os três reconhecedores perdem 22-42%. Verificando cada palavra com mais duas famílias de reconhecedores reduz o erro TTS aparente em palavras que pelo menos um reconhecedor ouve de duas a três vezes (muito raras palavras do LibriSpeech: de 10-13% a 3-5%), mas palavras muito raras permanecem as mais difíceis. Ig

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