System voices
Generate audio from text in the conversation, using the organization’s saved voices.
Changelog: v2026.10.11 · v2026.10.9 · v2026.8.24
What it is for
- Send audio in the brand voice without recording on the microphone
- Keep tone, speed, and provider consistent across attendants
- Store the script on the same message, as a transcription
- Reuse the same voice on AI Agent audio replies
Where to find it
| Area | Path |
|---|---|
| Save voices | Sidebar → Voices |
| Use in chat | Message field → Record from text |
| Use on the AI Agent | AI Agent → Sending tab → Reply with audio |
The chat button only appears if there is at least one active voice and the channel accepts audio.
Interflow catalog
Voices has two tabs:
| Tab | What it does |
|---|---|
| Catalog | ElevenLabs (search, language, gender, age, and category), Gemini (30 voices), and OpenAI (11 voices). The free ElevenLabs sample does not spend credits. Test text generates an MP3 and lets you change the model. OpenAI and Gemini use the Interflow balance |
| My voices | Organization presets. Search and New voice live on this tab |
Use this voice saves the preset under My voices, available in chat and on the AI Agent. The starting model is Flash v2.5. The form shows each model’s price per 1K characters.
Generating audio with a catalog voice spends AI credits:
| Models | Price |
|---|---|
| Flash v2.5, v4 Turbo, and v3 Conversational | US$ 0.13 per 1K characters |
| v4, v3, and Multilingual v2 | US$ 0.26 per 1K characters |
Without balance, audio is not generated. Voices tied to your own integration still use that key and do not spend credits. MiniMax is not in the catalog yet.
Gemini
On the Gemini tab, Test text opens the filled-in text and the model. You can change the model there before generating. The default is Gemini 3.8 Flash-Lite TTS, including when you add the voice.
Under My voices and in the form, Test text uses the saved or in-progress model, voice, and other settings. The same test exists for ElevenLabs, OpenAI, and Minimax.
The file is MP3. The Gemini API returns PCM or WAV; Interflow converts it before storing.
Prerequisite
Add a TTS integration in Settings:
| Provider | Use |
|---|---|
| OpenAI | OpenAI speech voices and models |
| ElevenLabs | Voices from the ElevenLabs account |
| Minimax | Voices and language from the Minimax account |
| Gemini | Built-in voices, TTS model, and a text style. The same key as the Gemini integration |
These are the same providers as the flow audio nodes.
Create a voice
Admin and owner create, edit, and delete. Other members only use active voices.
- Open Voices → New voice
- Give it a name (e.g. Female attendant)
- Choose the integration — the provider comes from it
- Adjust voice, model, speed, and the other provider fields
- Leave it Active and save
An inactive voice disappears from the chat button, but stays in the list so you can turn it back on later.
To copy a voice, use Duplicate on the card or in the edit form. It opens a new voice with the same settings and a (copy) name — adjust and save.
Use it in a conversation
- Type the text in the message field
- Click Record from text
- One active voice: generates immediately
- Two or more: pick the voice in the menu
- The audio appears as a preview; the script collapses
- Edit opens the text again; Regenerate creates another file with the same voice
- Send
Removing the preview deletes the generated file. Regenerating also replaces the previous file.
What the customer receives
In Interflow the message is one: audio + text as transcription.
On WhatsApp (and voice channels without an audio caption), the customer only hears the audio. The script is not sent as a separate text message.
Limitations
- With no active voice, the button does not appear
- With no text in the field, it does not generate
- Channels that do not send audio do not show the button
- Only admin and owner change the voice catalog