Home Assistant: Whisper API Integration for Speech-to-Text

Integration works for Assist pipelines.

Requirements:

• Installed Whisper.cpp server in network

Server setup (whisper.cpp):

1. Git pull whisper.cpp
2. Build the server (example for CUDA):

cmake -B build -DGGML_CUDA=1
cmake --build build --config Release

3. Run the server:

./ai/whisper.cpp/build/bin/whisper-server -m /ai/models/whisper/ggml-large-v3-turbo-q8_0.bin --host 192.168.0.55 --port 5045 -l ru  -sow -sns --vad --vad-model /ai/models/whisper/ggml-silero-v6.2.0.bin --inference-path /v1/audio/transcriptions

Configuration:

Add to your configuration.yaml:

stt:
  - platform: whisper_api_stt
    server_url: "http://192.168.0.55:5045/v1/audio/transcriptions"
    model: "whisper-1"
    language: "ru-RU"
    temperature: 0.0

Parameters:

• server_url (Optional): URL of your whisper.cpp or OpenAI-compatible server. Defaults to OpenAI API.
• api_key (Optional): API key if required by your server.
• model (Optional): Model name. Defaults to whisper-1.
• language (Optional): Language code (e.g., en-US, ru-RU).
• temperature (Optional): Sampling temperature between 0 and 1. Defaults to 0.0.
• prompt (Optional): Optional text to guide the model's style.

Notes:

• The integration converts the language code to ISO-639-1 (e.g., ru-RU -> ru) for API compatibility.
• Ensure your server_url includes the full path to the endpoint, e.g., /v1/audio/transcriptions.

Used sources + thanks to:

• sfortis/openai_tts: https://github.com/sfortis/openai_tts