Voice, speech and talking avatars
Adding text-to-speech, voice input and talking avatars to the Aimogen Pro chatbot, across OpenAI, ElevenLabs, Google, Azure and D-ID.
The chatbot can speak its replies, listen to spoken input, and render a talking avatar.
AI Chatbot › Text-to-Speech/Video
Provider credentials are covered in Speech and voice providers.
Enabling speech#
| Setting | What it does |
|---|---|
Enable Chatbot Text-to-Speech/Video | Master switch, and selects the provider |
Show Text-To-Speech Mute Button | Lets visitors mute |
Enable Visual Waveform Animation When The Chatbot Is Speaking | Animation during speech |
Waveform Animation Color | Default violet |
Voice input#
| Setting | What it does |
|---|---|
Enable Chatbot Voice Input | Adds a microphone button |
Enable Chat Auto-Submit On Voice Input Completion | Sends automatically when the visitor stops speaking |
Voice Button Color / Voice Button Activated Color | Styling |
Auto-submit is convenient but unforgiving — a pause mid-sentence sends a partial message. Leave it off unless you have tested it with real users.
Voice input uses the browser speech recognition API, so availability and quality depend on the browser.
Providers#
OpenAI#
| Setting | What it does |
|---|---|
Voice AI Model | The speech model |
AI Voice Selector | Which voice |
AI Voice Output Format | Audio format |
Voice Stability | Rendering speed and consistency |
Simplest option if you already use OpenAI — no extra account.
ElevenLabs#
| Setting | What it does |
|---|---|
Select a Voice | From your voices plus the built-in list |
Custom Voice ID | A cloned or custom voice by ID |
Voice AI Model | Which ElevenLabs model |
Voice Stability | Low gives more expression and more variance; high is flatter and more consistent |
Voice Similarity Boost | Closeness to the reference voice. Very high can amplify source artefacts |
Style Exaggeration | Amplifies character, costs latency |
Speaker Boost | Improves similarity at some processing cost |
Generally the best-sounding for English.
Google Text-to-Speech#
| Setting | What it does |
|---|---|
Voice Language | Locale, which filters the voice list |
Voice Name | A specific voice |
Voice Model | Standard, WaveNet, Neural2 |
Audio Device Profile | Optimise for headphones, phone speaker and so on |
Voice Speed / Voice Pitch | Delivery |
Widest language coverage.
Azure Speech Services#
Powers plain speech and the rendered avatar.
| Setting | What it does |
|---|---|
Azure API Key Region | Must match your Speech resource region |
Chatbot Avatar Character / Character Style | Which avatar, and its pose |
Chatbot Avatar Canvas Width | Rendered width |
Chatbot Voice | Neural voice |
Chatbot Voice Profile ID | Optional custom profile |
Private Azure Speech Resource Endpoint URL | Optional private endpoint |
Custom Voice Deployment ID | Optional custom voice deployment |
D-ID#
Animates a still photograph so it appears to speak.
| Setting | What it does |
|---|---|
Actor Source Image URL | The face to animate. A clear front-facing portrait works best |
Select a Voice | The D-ID voice |
AI Talking Avatar Width / Height | Rendered size |
This tutorial was recorded before Aiomatic was renamed to Aimogen Pro. Some labels or interface elements may differ slightly from the current version.
This tutorial was recorded before Aiomatic was renamed to Aimogen Pro. Some labels or interface elements may differ slightly from the current version.
Cost#
Speech is billed per character, which makes it far more expensive than it looks. A 200-word reply is around 1,200 characters, and a busy chatbot produces thousands of replies.
System & Logs › Text-to-Speech Limits provides a separate credit pool with its own per-user and per-guest allowances. Configure it before enabling speech publicly.
Talking avatars are more expensive again — video rendering is billed per generation.
Accessibility#
Speech output helps some visitors and hinders others.
- Always provide the mute button. Enable
Show Text-To-Speech Mute Button. - Do not autoplay audio. Unexpected sound is disorienting for screen reader users.
- Keep the text. Speech should accompany the written reply, never replace it.
Common problems#
No voices listed. The provider key is missing or invalid. Re-save the API key.
Audio cuts off. The reply exceeded the provider per-request character limit. Shorten replies with
Chatbot Max Token Count.
The avatar does not render. Azure free tier, wrong region, or an ad blocker.
Voice input does nothing. The browser does not support the speech API, or the page is not served over HTTPS — most browsers require a secure context for microphone access.
Related#
- Speech and voice providers
- Realtime voice chatbot — true speech-to-speech
- Audio, speech and transcription
Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.