Выпуск
Why Speech Transducers Omit Common Symbols
My Weird Prompts · 9 ноября 2026 г. · 25 мин
When speech models fail to produce common characters like email at-signs, the limitation stems from deep architectural tradeoffs rather than missing training data. RNN-T and TDT architectures calculate a four-dimensional joint tensor that scales directly with vocabulary size, making large vocabularies memory-prohibitive. This architectural constraint separates models into two camps: massive LLM-style decoders like Whisper that emit symbols directly, and compact transducers like Parakeet that rely on external inverse text normalization layers to format raw spoken words downstream. Episode 6508 — myweirdprompts.com/n/6508 (permanent short link: myweirdprompts.com/227172) Listen on Android: MWP FM Player — myweirdprompts.com/apk