Download the Right Model(s)

Download the tiny.en language model 

→ this is the only model that will provide useful results on a Raspberry Pi 500

bash ~/whisper.cpp/models/download-ggml-model.sh tiny.en

Better hardware will support bigger language models. 

Language Models: What is the difference? 

Whisper.cpp provides several different language models. Execute the script ~/whisper.cpp/models/download-ggml-model.sh without parameters to see currently available models^#: 

mela@Cox:~/meetings/recordings/2026-02-21T013852 $ ~/whisper.cpp/models/download-ggml-model.sh
Usage: /home/mela/whisper.cpp/models/download-ggml-model.sh <model> [models_path]

Available models:
  tiny tiny.en tiny-q5_1 tiny.en-q5_1 tiny-q8_0
  base base.en base-q5_1 base.en-q5_1 base-q8_0
  small small.en small.en-tdrz small-q5_1 small.en-q5_1 small-q8_0
  medium medium.en medium-q5_0 medium.en-q5_0 medium-q8_0
  large-v1 large-v2 large-v2-q5_0 large-v2-q8_0 large-v3 large-v3-q5_0 large-v3-turbo large-v3-turbo-q5_0 large-v3-turbo-q8_0

___________________________________________________________
.en = english-only -q5_[01] = quantized -tdrz = tinydiarize

Models that only support English are least demanding of resources, especially memory. Multilingual models provide support for 99 different languages, and contain, as such, necessarily some overhead. 

While multilingual models from tiny to large all support the same set of languages, the size of the model determines how well Whisper.cpp handles accents, mumbling, people talking over each other or specialist language. 

Since this describes the typical TTRPG environment pretty well, smaller language models — while running on inexpensive hardware — will provide only limited results. 

^#) for an explanation of quantization and diarization look here