Voice and microphone
Choose the character’s speaking voice, test it and understand how browser microphone input becomes a message.
Choose and preview a voice
- Open the character from Characters (Open studio), then choose Language & voice in the right dock. Its tab is Language & Voice.
- Select a TTS Provider: ElevenLabs (the included IAMX voice or your own account), MiniMax, OpenAI TTS or Free TTS (included voices, no API key required). MiniMax and OpenAI TTS need a key: a workspace key saved in Settings is used automatically (Override for this character sets a separate one); without one, enter the key on the tab. For MiniMax, Fetch voices loads its voices; OpenAI TTS has no voice list, so open Enter voice ID manually and type a voice name such as
alloy,echoornova. - For ElevenLabs, the Voice connection card says which connection speaks: “Using your account key”, “Using this character’s saved key” or “Using the included IAMX voice”. Manage my key in Settings opens the workspace keys.
- Choose a voice in Voice Selection. The list depends on the provider and the key. Refresh voices (ElevenLabs and Free TTS) reloads it, Enter voice ID manually takes a voice ID directly where offered, and Free TTS also offers Automatic (match language).
- For ElevenLabs and OpenAI TTS, choose a TTS Model. For ElevenLabs, the tab starts on Flash v2.5, recommended for live conversation; Eleven v4 Turbo is the newest real-time model (about 150 ms to the first audio), and Eleven v4 has the highest quality but is slower in live conversation (about 1 s). With ElevenLabs, also pick a Voice Delivery profile. Adjust the Prosody sliders, then press Preview selected voice and settings.
- Press Save (Ctrl/Cmd+S), then open Test conversation in the right dock to hear a real reply on the stage. A voice preview does not test the AI model or your knowledge sources.
The languages a character answers in are set on its Identity tab, under Languages (All languages or Selected languages), not on Language & Voice. Which spoken languages work also depends on the voice and the speech recognition.
Create your own voice
With ElevenLabs selected, Create your voice (marked “Starter and above”) turns a recording into a voice for your characters. Enter a Voice name, add an Audio recording (MP3, WAV, M4A, OGG or WebM, up to 20 MB, one speaker, no background music), confirm that it is your voice or that you have the speaker’s permission, and press Create voice. On Free, the card offers View paid plans instead.
Without your own ElevenLabs key, the voice is created for your workspace and counts toward its personal voice limit. With your own key, it is added to your ElevenLabs account, and your provider’s limits and charges apply. A voice that needs provider verification is listed under Voices awaiting verification until it can be used. Delete my voice removes a personal voice you have selected.
Voice clones created before v4 must be retrained in ElevenLabs before they work with Eleven v4 or Eleven v4 Turbo. The TTS Model card shows this reminder when you choose a v4 model with a cloned or personal voice selected; until the clone is retrained, keep an earlier model such as Flash v2.5 for that voice.
Realtime voices
When the character’s Core AI uses OpenAI Realtime (voice) or Gemini Live (realtime), the voice comes directly from that model. Language & Voice then shows only OpenAI Realtime Voice or Gemini Live Voice with that model’s own voices; ElevenLabs, MiniMax and the other TTS providers are not used. These voices have no preview button: listen to them in Test conversation. ElevenLabs voice IDs and realtime voice names are not interchangeable, so revisit Language & Voice after changing Core AI. See AI models.
Speak in the browser
Use HTTPS and allow the microphone when the browser asks. In the web conversation — the embed, the website widget and the editor’s Test conversation — one microphone button works two ways (the embed’s hint: “Tap: hands-free · Hold: push-to-talk”):
- Tap turns hands-free listening on; each phrase is sent when you stop speaking. Tap again to turn it off.
- Hold the button and talk; the phrase is sent when you release it.
In the standard mode, the conversation uses the browser’s own speech recognition where the browser supports it; otherwise the audio is sent to IAMX’s speech recognition. Hands-free listening pauses while the character answers; hold the button to interrupt. With a realtime model the audio always goes to the model, and hands-free keeps listening while the character answers, so you can talk over it. Browser support, permissions and connectivity vary, and browser dictation does not work offline.
The kiosk page shows “Tap the microphone to talk”. A Pixel Streaming character listens through the browser’s speech recognition only; in a browser without it, type your message instead.
For an Unreal client, the IAMX component’s own Interaction Mode decides how a turn starts; see the IAMX component. The character’s Identity tab also has an Interaction Mode (Manual (text/button), Proximity, Voice-activated, Push-to-talk or Wake word), saved with the character. With Wake word, the wake phrases are checked on the server, so Unreal, web and kiosk follow the same rule.
If the result is wrong
| Symptom | What to check |
|---|---|
| No voice or only one voice appears | Check the selected provider and key, voice access and provider permissions, then reload the list (Refresh voices, or Fetch voices for a key entered on the tab). “Could not load the voice library” points to the key or connection in Settings. A saved voice ID may still appear even if the list could not be loaded. |
| “Voice connection unavailable” | No included voice is available for this connection. Add your own provider key in Settings, or contact your administrator. |
| Text arrives but no sound | Tap inside the experience to allow audio, check the sound control and run the voice preview. Read any provider quota or key error. If the saved voice is no longer allowed, the character falls back to the included default voice; if no voice can be used at all, it answers in text only. |
| “Microphone access was denied” | Allow the microphone for the site in your browser settings, then tap the microphone again. |
| “Browser voice input could not start” | Tap the microphone to try again. If “The microphone is unavailable” appears, check that a microphone is connected and selected. |
| “Voice input is unavailable in this browser” | The browser has no speech recognition for this character. Type your message, or use a browser that supports it. |
| Microphone keeps listening | Tap the button to turn hands-free off, or use hold-to-talk. Check the selected microphone and the on-screen message before retrying. |
| Words are misheard | Check the character’s Languages on the Identity tab, reduce background noise and use headphones. Compare a typed question with the recognised phrase in Insights › Sessions. |
| Replies feel slow | Compare typed and spoken questions. Turn detection, transcription, model generation, tools, speech synthesis and network time all add up; a provider’s advertised speech latency is not the total response time. For ElevenLabs, a faster TTS Model such as Flash v2.5 (about 75 ms) or Eleven v4 Turbo (about 150 ms) shortens the voice step; Eleven v4 takes about a second to start speaking. |
Share useful diagnostics
For support, provide the browser, device, time, character and the exact error text. Do not paste an API key or someone’s private conversation. See the troubleshooting checklist and Support.
Updated . Your available options depend on the character, plan and account.