Setting Up Faster-Whisper for Speech Recognition

LocalRPG supports voice dictation, allowing you to speak your character's actions and dialogue directly into your microphone. Rather than sending microphone audio to third-party cloud services, you can run a local OpenAI-compatible transcription server powered by faster-whisper (CTranslate2).

1. Hardware Requirements

Faster-Whisper is optimized for high-speed inference on both CPUs and GPUs:

Model	Memory (VRAM / RAM)	Accuracy	Relative Speed
tiny / base	< 1 GB	Basic commands	~16x real-time on GPU, ~3x on CPU
small (Recommended)	~1.5 GB	Good for RPG prose & character dialogue	~10x real-time on GPU, ~1.5x on CPU
medium / large-v3	~3–5 GB	Near-perfect multilingual transcription	~5x real-time on GPU

[!TIP] The small or base.en models provide instantaneous transcription and rarely mishear fantasy terms.

2. Running via Docker (Recommended)

The community-standard fedirz/faster-whisper-server provides an OpenAI-compatible /v1/audio/transcriptions endpoint on port 8000.

GPU Acceleration (NVIDIA CUDA)

docker run -d \
  --name faster-whisper \
  --restart always \
  --gpus all \
  -p 8000:8000 \
  -e WHISPER_MODEL=Systran/faster-whisper-small \
  fedirz/faster-whisper-server:latest-cuda

CPU Execution (No GPU required)

docker run -d \
  --name faster-whisper \
  --restart always \
  -p 8000:8000 \
  -e WHISPER_MODEL=Systran/faster-whisper-small \
  fedirz/faster-whisper-server:latest-cpu

Docker Compose (docker-compose.yml)

services:
  whisper:
    image: fedirz/faster-whisper-server:latest-cuda # or latest-cpu
    container_name: faster-whisper
    restart: always
    ports:
      - "8000:8000"
    environment:
      - WHISPER_MODEL=Systran/faster-whisper-small
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

3. Verify Server Health

Test transcribing an audio file:

curl -s -X POST http://localhost:8000/v1/audio/transcriptions \
  -H "Content-Type: multipart/form-data" \
  -F file=@test.wav \
  -F model="whisper-1"

Verify that the returned JSON contains the transcription text:

{"text":"Greetings, adventurer! The local speech engine is online and ready."}

4. Connecting in LocalRPG

Via Settings Studio (GUI)

1. Open Settings Studio -> Media -> STT.
2. Click Load STT Preset… and select Faster-Whisper.
3. Confirm the Endpoint is set to http://localhost:8000/v1/audio/transcriptions and Model to whisper-1.
4. Click Save Settings.
5. When playing a campaign, click the microphone button next to the prompt bar to speak your action.

Via Configuration File (config.yaml or localrpg.yaml)

media:
  stt:
    type: http
    endpoint: "http://localhost:8000/v1/audio/transcriptions"
    model: "whisper-1"

5. Alternatives: Whisper.cpp (whisper-cli)

If you prefer a standalone command-line binary without a long-running HTTP server, LocalRPG also supports whisper-cli from the whisper.cpp project:

media:
  stt:
    type: cli
    command: "whisper-cli"
    args:
      - "-m"
      - "models/ggml-base.bin"
      - "-f"
      - "%INPUT%"
      - "-nt"