Official documentation
speech.cpp: docs/models.md (model catalog, tasks and languages)
nyosegawa, GitHub, nyosegawa/speech.cpp, 2026
Read the originalRead on October 22, 2026
- What it is
- The model catalog in the documentation of speech.cpp, a speech program written in C++ on ggml. A table lists each model with its task and its languages, including the recognition models Qwen3-ASR, Parakeet TDT and ReazonSpeech.
- How much weight it carries
- Written by the project's author in the project's own repository, so it states what speech.cpp declares it supports, not an independent test. The counts are declared languages per model and say nothing about accuracy in each language.
Facts taken from this source
- The speech.cpp model catalog lists one language, Japanese (ja), for the recognition model parakeet-tdt_ctc-0.6b-ja.
“Name Task Languages ... parakeet-tdt_ctc-0.6b-ja recognition ja” (Model table in docs/models.md, row parakeet-tdt_ctc-0.6b-ja)
- The speech.cpp model catalog lists 25 European languages for the recognition model parakeet-tdt-0.6b-v3.
“Name Task Languages ... parakeet-tdt-0.6b-v3 recognition 25 European languages” (Model table in docs/models.md, row parakeet-tdt-0.6b-v3)
- The speech.cpp model catalog lists 30 languages for the recognition model qwen3-asr-0.6b.
“Name Task Languages ... qwen3-asr-0.6b recognition 30 languages” (Model table in docs/models.md, row qwen3-asr-0.6b)
- The speech.cpp model catalog lists 30 languages for the recognition model qwen3-asr-1.7b.
“Name Task Languages ... qwen3-asr-1.7b recognition 30 languages” (Model table in docs/models.md, row qwen3-asr-1.7b)
- The speech.cpp model catalog lists one language, Japanese (ja), for the recognition model reazonspeech-v2.
“Name Task Languages ... reazonspeech-v2 recognition ja” (Model table in docs/models.md, row reazonspeech-v2)