# Voices Everything about how the agent sounds is on **Identity - Voice & Language**. Each language has its own voice: the primary one under **Primary Language & Voice**, the others under **Secondary Languages**. Changes apply from the next call. ## Pick a voice 1. **Open the agent - Identity - Voice & Language** **Preview** plays the current voice. 2. **Click Change voice** The **Choose a voice** window opens on the provider of the current voice. Search by name, or press **Play sample**. 3. **Hear it say your own words** Type a sentence your callers will really hear under **Test voices with custom text**, such as your greeting, a price or an address, then click **Generate preview** on a voice. Samples in the catalog rarely include Greek numbers and names. 4. **Select it, save, and make a test call** A voice sounds different on a phone line than in the browser. Judge it on a real call. ## Which provider - **The most natural voice and every setting** — turn on **ElevenLabs (default)** (needs Voice model choice (Expressive or Most options), stability, similarity, style and speaker boost.) - **A different voice character** — turn on **Grok or Gemini** (needs Only speech speed can be tuned. Gemini voices take longer to start speaking: expect about 1.5-2 seconds of extra pause before each reply.) - **Your own voice, or your receptionist's** — turn on **Cloned voices** (needs Record or upload a sample. See [Clone a voice](#clone-a-voice).) Every voice on these providers speaks all supported languages, without a foreign accent. To keep one voice across languages, use the same voice for each language. Below the voice, **Callers can also hear ... in:** lists the languages that voice already speaks. ## Voice model (ElevenLabs) Under **Advanced - Voice tuning**, **Voice model** sets how an ElevenLabs voice speaks: - **Expressive** — The most human-sounding voice. It shapes speed, style and emotion by itself, from the text and [audio tags](#audio-tags), so you set only **Voice stability** (around 0.5 suits most calls) and **Voice similarity**. Best for reception, bookings and complaints, where warmth matters. - **Most options** — The fastest replies and every voice setting: speed, stability, similarity, style and speaker boost. Best when every tenth of a second counts, or when you need a fixed speaking speed. - **Platform default** — Follows the platform setting, which may change over time. The label shows which model it is today. After switching, make a test call and listen to a whole conversation, not just the greeting. ## Voice tuning - **Speech speed** — 0.7 (slower, clearer) to 1.2 (faster). 1.0 is normal. Not shown with Expressive, which sets its own pace. - **Voice stability** — Lower: more emotional and varied. Higher: steadier and more consistent. If the voice changes tone between sentences, raise it. - **Voice similarity** — How close the voice stays to its original recording. Higher: closer to the original, slightly less natural. - **Style exaggeration** — Strengthens the voice's own style. Keep it at 0 for the shortest pauses. - **Speaker boost** — Closer to the original speaker, at a small extra delay. - **Allow caller to request speed change** — Callers can ask the agent to speak faster or slower during the call. - **Auto-adjust speed based on caller** — The agent matches the caller's speaking pace. Stability, similarity, style and speaker boost are for ElevenLabs voices only. ## Audio tags With the **Expressive** voice model, short English directions in square brackets change how the next words sound, for example `[warmly]`, `[sympathetic]`, `[cheerfully]` or `[whispers]`. They are never spoken and never appear in transcripts. - **Automatically:** the agent adds them itself when the caller's words carry a feeling, such as `[sympathetic]` for a complaint. - **Welcome and fixed messages:** write them at the start of a sentence, e.g. `[warmly] Welcome to our clinic!` - **Prompt:** tell the agent when to use one, e.g. *"When the caller thanks you, start with [cheerfully]."* - **Not in:** SMS, emails, or sentences with names, numbers, dates and prices. Those stay plain and clear. One tag per sentence works best. Tags that describe the voice work better than sounds. An ellipsis (…) adds a pause, and CAPITALS add emphasis. ## Languages Under **Secondary Languages**, **Add language** gives the agent another language with its own voice. Pick **Same voice as main language** to keep one voice for all of them. More on language switching: [Your first agent](https://docs.voicelogica.ai/getting-started/your-first-agent/#identity---voice--language). To change the language the agent starts every call in, pick another **Primary language** under **Primary Language & Voice**. The agent keeps its voice when that voice speaks the new language, otherwise it picks one that does. The welcome message and the prompt stay as they are written, so rewrite them in the new language. ## Background sound **Call Background Audio** plays a quiet ambience under the whole call: **Office**, **Call center** or **Coffee shop**. Keep the volume low (about 10-20%). Louder ambience covers the voice and makes the agent mishear callers. There is no music under the conversation; music plays only while a call is being transferred. ## Clone a voice 1. **Change voice - Cloned voices - Clone a voice** The **Cloned voices** tab shows when your plan includes voice cloning, or when you use your own ElevenLabs key. 2. **Name it and pick the provider** For example **Maria - reception**. ElevenLabs or Soniox. 3. **Record or upload one speaker** Clear audio, no music, nobody else talking. 30 seconds to 2 minutes works best (Soniox uses the first 20 seconds). Files must be under 25 MB. 4. **Click Clone voice, then select it** The clone speaks every language, not only the language of the recording. Clone only a voice you have permission to use. **Delete voice** removes it from the provider too, and cannot be undone. A voice that an agent still uses cannot be deleted. ## With Claude or ChatGPT With the [Voice Logica connector](https://docs.voicelogica.ai/integrations/claude-chatgpt/), you can ask in your own words, for example: *"show me Greek female voices"*, *"switch my agent to the Expressive voice"*, *"make it speak a little slower"*, *"make English the main language and keep Greek as a second one"* or *"clone this recording as Maria"*. The assistant lists the voices with sample links and makes the change for you. ## Troubleshooting **Callers hear a pause before every answer** Check the provider first: Gemini voices add about 1.5-2 seconds before each reply. Switch to an ElevenLabs voice. With ElevenLabs, **Most options** replies fastest, and **Style exaggeration** at 0 and **Speaker boost** off save a little more. A pause only after certain questions usually comes from the agent looking something up, not from the voice. **The voice sounds flat or robotic** Pick an ElevenLabs voice, set **Voice model** to **Expressive** and **Voice stability** to about 0.5. Test with your own sentences under **Test voices with custom text** before you choose. **The greeting says «η ψηφιακή βοηθός» but the voice is male** The greeting and the prompt speak as one gender, and the voice should match it. Either pick a voice of that gender, or change the wording of the greeting (**ο ψηφιακός βοηθός** for a male voice). **A provider tab says Not Enabled** That provider has no API key on your account. Use another provider, or add your key under **Manage API keys**.