Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

I recorded 1,180 TTS clips to find out what voices misread in LLM output

Описание и применение

Предобработка текста LLM с помощью библиотеки speechready снижает количество ошибок чтения голосовыми агентами с 32% до 8% в общей сложности, а также в английском и турецком языках, без ухудшения контрольного текста.

Задача

Voice agents often misread text generated by LLMs, especially numbers, dates, phone numbers, and code, leading to incorrect spoken output.

Результат

Preprocessing with speechready reduced overall misread rate from 32% to 8%, with similar improvements in English and Turkish, and no regression in control prose.

Рабочий процесс

  • 1. Collect LLM-generated text containing markdown, numbers, dates, phone numbers, URLs, and code.
  • 2. Run speechready to clean, verbalize, compile, and chunk the text.
  • 3. Synthesize the original and processed text with each TTS voice.
  • 4. Transcribe both recordings with gpt-4o-transcribe.
  • 5. Score the transcriptions against the intended words, forgiving transcription conventions.
  • 6. Calculate misread rates and compare before/after.
  • 7. Iterate on rules and provider profiles based on held‑out tests.

Доказательства

I recorded 1,180 TTS clips to find out what voices misread in LLM output

Источник

Result, on the same 1,180 recordings: All 32% → 8%

Источник

The scorer forgives the ways a transcription engine writes speech down: digits for spoken numbers, "21st" for "twenty-first", "NPM" for "en pi em".

Источник

Providers disagree on what they handle. Google's Chirp voices read 25°C fine; others do not.

Источник

  • Дата записи: 2026-11-02T22:29:04.921Z