Home Catalog Speaches Speech-to-Text & Audio
AI & LLM Services

Speaches Speech-to-Text & Audio

Speaches is an open-source, OpenAI-compatible Speech-to-Text (faster-whisper) and Text-to-Speech (Kokoro/Piper) server supporting GPU acceleration.

ARM64 & AMD64
Docker & Rootless Podman
100% Self-Hosted (Zero Lock-in)
RAM Profile: Low
Default Port: Port 8008
Deploying on a home server? NjordDeploy automates the full setup, SSL, storage, and networking on your Raspberry Pi or Proxmox host.
Discover 128+ Apps

Why this Configuration?

Quick Start (Standalone Docker Compose)

The snippet below is immediately ready to run in any standard Docker or Podman environment:

services:
  speaches:
    container_name: njorddeploy-speaches
    image: "ghcr.io/speaches-ai/speaches:latest-cpu"
    volumes:
      - "./data/speaches/cache:/home/ubuntu/.cache/huggingface/hub"
    ports:
      - "8008:8000"
    environment:
      - "WHISPER__COMPUTE_TYPE=default"
      - 'PRELOAD_MODELS=["Systran/faster-whisper-small"]'
      - "API_KEY=speaches"
    networks:
      - njorddeploy_net
    restart: unless-stopped

networks:
  njorddeploy_net:

Start the service directly via the command line:

Docker Engine
docker compose up -d
Rootless Podman
podman compose up -d

Configuration & Environment Variables

Key configuration parameters and defaults derived from the NjordDeploy component template:

Variable Default Value Description
SPEACHES_PORT 8008 The host port to expose the Speaches STT/TTS API.
SPEACHES_MODEL Systran/faster-whisper-small Whisper spraakmodel voor transcriptie. 'Systran/faster-whisper-small' is de aanbevolen standaard voor uitstekend Nederlands op standaard CPU (~1 GB RAM).
SPEACHES_COMPUTE_TYPE default Whisper compute type (e.g. default, int8, float16, int8_float16).

Ecosystem & Enterprise Integration

Effortless Management with NjordDeploy

Deploy, monitor, and update this service with a single click on your own hardware via the NjordDeploy Configurator.