Speech and voice providers

Configuring text-to-speech, speech-to-text and talking-avatar providers in Aimogen Pro, including OpenAI, ElevenLabs, Google, Azure Speech and D-ID.

Aimogen Pro uses speech services in three places: reading chatbot replies aloud, generating audio versions of posts, and transcribing uploaded audio. Credentials live in Settings › API Keys; behaviour is configured per feature.

At a glance#

ProviderCredentialUsed for
OpenAImain OpenAI keyText-to-speech and speech-to-text
ElevenLabsElevenLabs.io API KeyText-to-speech, voice cloning
Google Text-to-SpeechGoogle Text-to-Speech API KeyText-to-speech
Azure Speech ServicesAzure Speech Services API KeyText-to-speech and talking avatars
D-IDD-ID API KeyTalking avatar video

OpenAI#

Uses the OpenAI key already configured.

Speech-to-text models: whisper-1, gpt-4o-mini-transcribe, gpt-4o-transcribe. This is the only transcription backend in the plugin, used by the audio converter shortcode, the speech-to-text OmniBlock and the AI Content Editor audio tools.

Text-to-speech offers a voice selector, an output format and a speed control, configured under AI Chatbot › Text-to-Speech/Video and in the AI Content Editor.

ElevenLabs#

Key: elevenlabs.ioSettings › API Keys › ElevenLabs.io API Key.

The plugin fetches your available voices and caches them in the aiomatic_elevenlabs option, merged with a built-in default list (Rachel, Domi, Bella, Antoni, Elli, Josh, Arnold, Adam, Sam).

Voice parameters, under AI Chatbot › Text-to-Speech/Video:

SettingWhat it does
Select a VoiceOne of your voices, or a stock voice
Custom Voice IDUse a cloned or custom voice by ID instead of the dropdown
Voice AI ModelWhich ElevenLabs model renders the audio
Voice StabilityLow values give more emotional range and more variance between renders; high values are more consistent and flatter
Voice Similarity BoostHow closely the output tracks the reference voice. Very high values can amplify artefacts from the source recording
Style ExaggerationAmplifies the voice stylistic character. Costs latency
Speaker BoostImproves similarity at some processing cost

ElevenLabs is generally the best-sounding of the four for English narration.

Google Text-to-Speech#

Key: a Google Cloud API key with the Cloud Text-to-Speech API enabled → Settings › API Keys › Google Text-to-Speech API Key.

Voices are fetched and cached in the aiomatic_google_voices option.

SettingWhat it does
Voice LanguageLanguage and locale, which filters the voice list
Voice NameA specific Google voice
Voice ModelStandard, WaveNet, Neural2 and similar tiers. Higher tiers sound better and cost more
Audio Device ProfileOptimises the audio for a playback device class, for example headphones or a phone speaker
Voice SpeedPlayback rate
Voice PitchPitch shift

Google has by far the widest language coverage of the four.

Azure Speech Services#

Key: an Azure Speech resource key → Settings › API Keys › Azure Speech Services API Key.

Azure powers both plain text-to-speech and the talking avatar feature, where a rendered character speaks the reply on a canvas.

SettingWhat it does
Azure API Key RegionMust match the region your Speech resource was created in
Chatbot Avatar CharacterWhich stock avatar to render
Chatbot Avatar Character StylePose and framing
Chatbot Avatar Canvas WidthRendered width in pixels
Chatbot VoiceThe Azure neural voice
Chatbot Voice Profile IDOptional custom voice profile
Private Azure Speech Resource Endpoint URLOptional private endpoint
Custom Voice Deployment IDOptional custom voice deployment

D-ID#

Key: studio.d-id.com account settingsSettings › API Keys › D-ID API Key.

D-ID animates a still photograph so it appears to speak the reply.

SettingWhat it does
Actor Source Image URLA photograph of the face to animate. A clear, front-facing portrait works best
Select a VoiceThe D-ID voice
AI Talking Avatar Width / HeightRendered size

D-ID supports a streaming mode for lower latency. See Voice, speech and talking avatars.

Usage limits#

Speech is metered separately from text. System & Logs › Text-to-Speech Limits provides its own credit pool with the same structure as the text limits: per-user and per-guest credits, a time frame, and an absolute or rolling window.

Configure this before exposing voice output to visitors. Speech is billed per character and adds up much faster than people expect. See Usage limits and credits.

Choosing#

GoalSuggestion
Best English narrationElevenLabs
Widest language coverageGoogle
Enterprise Azure estate, or a rendered avatarAzure Speech Services
Fewest extra accountsOpenAI
A photorealistic talking personD-ID

Common problems#

No voices in the dropdown The key is missing or wrong, so the voice list could not be fetched. Re-save the API key.

Azure avatar shows nothing Free tier, wrong region, or an ad blocker. All three are common.

Audio cuts off The text exceeded the provider per-request character limit. Split long content.

Costs higher than expected Speech is per character. Add a text-to-speech limit.

Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.