The Speech Pipeline
STT: Speech-to-Text
The entry point of the pipeline. It listens to the user’s voice and converts it into text the Agent can process. Your choice here affects how accurately the Agent understands the user, how quickly it responds, and which languages it handles. Two platforms are available: Deepgram and ElevenLabs.Deepgram
A real-time speech recognition platform built for low-latency, high-accuracy transcription. Nine models are offered, spanning three generations plus phone-call variants.ElevenLabs
ElevenLabs offers a single real-time transcription model in OpenHome.A note on short utterances
STT engines use a Voice Activity Detector (VAD) to decide when you have finished speaking. The engine listens for a silence gap after your voice, and only once that gap is detected does it finalize the transcription and pass it to the Agent. When you say a single short word, the engine may not detect a clean silence gap quickly, especially with background noise, so the response can feel delayed. Speaking in short phrases produces faster, more reliable results.General interaction
Instead of saying a single word and waiting, speak a short, complete phrase:Wake word
The same applies when using a wake word. Saying it alone and pausing can cause a slow response, because the engine waits for more audio to confirm the utterance is complete. Pair the wake word with a short phrase.
See Wake Word and Sleep Interaction for more on how this affects the wake word flow.
Music mode
In music mode, background audio fills the silence the VAD is listening for, making single-word commands harder to detect. If you say just “stop” or “pause” while music is playing, the engine may not cleanly finalize the transcription before more audio arrives. Add a word before or after the command to form a short phrase:TTT: Text-to-Text
The Agent’s brain. Once speech has been transcribed, the text goes to a language model that generates the Agent’s response. Your choice here shapes how well the Agent understands context and how quickly it replies. Three platforms are available: OpenAI, OpenRouter, and Venice. Each needs its own API key under Settings → API Keys.OpenAI
Direct access to OpenAI’s models using your own OpenAI API key.gpt-4o-mini is the small, fast, low-cost option. The two GPT-5.6 models are the larger, more capable ones, and are the better choice when the Agent needs to follow a long or subtle prompt.
OpenRouter
A unified gateway that reaches models from several vendors through a single API key, so you can switch provider without managing separate keys.
For a voice Agent, favour the models marketed as fast or small, such as
openai/gpt-5-nano or google/gemini-3-flash-preview. Latency is felt directly in conversation in a way it is not in a chat window.
Venice
Venice is a privacy-first, OpenAI-compatible gateway that does not retain your data, and it carries by far the widest model catalogue of the three. It includes open-weight models alongside frontier ones, and some deliberately unrestricted models.The
venice-uncensored-* and gemma-4-uncensored models apply far fewer content restrictions than the others. Choose them deliberately, and only where that behavior is appropriate for the Agent and its audience.LLM fine-tuning parameters
These sit below the model selectors in Settings → Configuration and apply to the selected Text-to-Text model. The UI labels them with anOPENAI TTT prefix.
TTS: Text-to-Speech
The Agent’s voice. Once the language model has generated a response, the TTS stage converts that text into audio. ElevenLabs is the only TTS platform, using your own ElevenLabs API key.ElevenLabs lists both Turbo models as deprecated and recommends the Flash models in their place. For a new Agent, start with
eleven_flash_v2_5.Voice fine-tuning parameters
Per-Agent Voice Configuration
The configuration in Settings → Configuration applies across your account. When you need one Agent to sound or think differently, set it on that Agent instead. Open the Agent in the workspace, either by creating it or by choosing Edit from its More actions menu, and use the Voice & Speech section:- Voice: search the voice list or pick from Suggested for you. Each card has a play button to preview it.
- Create Voice: build a new voice without leaving the workspace.
- Gender: shapes the Agent’s interaction style.
Speech-to-Text and Text-to-Speech are account-wide only. The Agent workspace has no STT or TTS selector, so those two stages are always taken from Settings → Configuration. Only the LLM and the voice can be set per Agent.
Adding a Custom Voice
Custom voices live on the Voices page. Click Create Voice and pick how you want to build it.
Getting a Voice ID
If you are bringing a voice you already have, upload it to your TTS provider first (ElevenLabs, for example). The provider gives you a Voice ID, which you supply when creating the voice in OpenHome.Other Configuration Options
The rest of Settings → Configuration, in the order it appears on screen.API Keys for Providers
Each provider needs its own API key. Manage them under Settings → API Keys.When you update API keys, your Agents consume credits from the associated services. OpenHome is not responsible for charges from third-party providers.See Dashboard for full API key management details.
See also
- Configuring Your Agent: conversation controls, identity, and behavior prompts
- Wake Word and Sleep Interaction: how the wake word interacts with STT and the short-utterance pattern
- SDK Reference: OpenRouter model table for use inside Abilities