Uma única pesquisa de apt-cache search speech to text retornou um programa chamado "Julius":
Descrição do pacote julius :
speech recognition engine
Julius is a high-performance, two-pass large vocabulary continuous
speech recognition (LVCSR) engine.
.
It supports N-gram based dictation, DFA grammar based parsing, and
one-pass isolated word recognition. Phone context dependencies are
supported up to triphone. It can perform a multi-model decoding, a
recognition using several LMs and AMs simultaneously with a single
processor, and also supports "hot plugging" of arbitrary modules at
run time.
.
This package contains the executable applications, a set of tools
useful to build recognition grammar and some examples (like a
script to use Julius to execute some predetermined commands).
Instalação: sudo apt-get install julius