Voice & language

Choose the voice your callers hear and the language they hear it in - every option explained in plain language.

The voice and language are the very first thing a customer experiences on the call - before a single word of your carefully-written script matters. In the first two seconds someone decides whether they’re talking to a person who respects them or a robot they want off the line, and that snap judgment is made almost entirely on how the agent sounds, not what it says. Get this right and everything else you built has a chance to land. Get it wrong - a stiff voice, or the wrong language - and callers hang up before you’ve made your point.

The good news: it’s two settings, and you can hear both before you commit. You set them in Step 1 (Basics & voice) of the agent builder, in the Set up voice section. This page explains what each one does and how to choose well.

The one rule that matters most: match the voice and language to your callers, not to your team. The agent should sound like someone your customer would happily pick up the phone to - in the language they actually speak at home.

Choosing a voice

The voice (also called the speaker) is the actual person your caller hears. oyehello offers a large catalog of natural, low-latency voices - friendly names like Shubh, Priya, Aditya, Neha, Rohan, and many more. Shubh is the default on a new agent.

You pick one in the Set up voice grid. Each voice shows up as a card; the one you’ve chosen gets a highlight and a checkmark.

Every voice is multilingual

This is the important thing to understand, and it’s easy to miss: you don’t pick a “Hindi voice” or a “Tamil voice.” Every single voice in the catalog speaks all the supported languages. Shubh can open your call in Hindi, switch to English, and handle Tamil - same voice, same personality, whichever language you set. So choose the voice for its character and warmth, and set the language separately (next section).

Previewing before you commit

Never publish a voice you haven’t heard. On each voice card there’s a Preview button for every language you’ve chosen (one for your primary language, and a second if you set a secondary). Click it and you’ll hear the real production voice saying a sample line - not a demo clip, the actual voice that will be on the call.

  • How to use it: shortlist two or three voices, then preview the same one in the language your callers speak. Trust your ear - the voice that makes you want to keep listening is the one your customers will too.
  • The catalog is long. Use the Search voices box to jump to a name, or hit See all voices to browse the full set.

If the preview doesn’t play: it just means the preview service is briefly unavailable - it doesn’t affect your live calls. Try again in a moment.

Picking a voice that fits

There’s no “tone” or “style” label on each voice - you judge them by ear, which is exactly why previewing matters. A few practical instincts:

  • Warmer, everyday voices beat formal ones on the phone. People are wary of cold calls; a voice that sounds like a helpful neighbour disarms that instinct. Save the crisp, businesslike voices for reminders and notices where warmth isn’t the point.
  • Keep the voice consistent with the persona you wrote. If your agent’s purpose is gentle payment reminders, a brisk hard-sell voice fights the script. The voice and the words should feel like the same person.
  • The agent even adapts its gendered speech to match the voice you pick - so in Hindi and other Indic languages, a voice that sounds male or female uses the matching first-person verb forms rather than sounding grammatically off. You don’t configure this; choosing the voice is enough.

The primary language

The primary language is the language the agent actually speaks on the call. It’s a required field, and it’s the single biggest lever on whether your callers stay on the line.

oyehello supports 11 languages - English plus ten Indian languages:

  • English (IN)
  • Hindi
  • Bengali
  • Gujarati
  • Kannada
  • Malayalam
  • Marathi
  • Odia
  • Punjabi
  • Tamil
  • Telugu

Whichever you pick, that’s the language the caller hears from the opening line onwards - including the mandatory spoken notices (“this is an AI assistant,” “this call is being recorded”), which are delivered in the caller’s language so they actually understand them. Your call script and persona can be written in plain English in the builder; the agent still speaks to the caller in the primary language you set here.

Speak your list’s language, not your office’s. This is the mistake that quietly kills campaigns. Your team writes and thinks in English, so English feels like the natural default - but if your contact list is small-town borrowers who speak Hindi, or shopkeepers in Tamil Nadu, an English opener makes you sound like an outsider they don’t trust. If your callers speak Hindi, set Hindi. The agent should meet them where they are.

What about Hinglish?

Hinglish - that natural mix of Hindi and English so many Indians actually speak - isn’t a separate option in the dropdown, but you get it easily: set the primary language to Hindi, then write your call context in natural Hinglish. Mix in the English words people really use (“aapka payment”, “EMI due hai”, “confirm kar dijiye”) and the agent mirrors your phrasing on the call. So you don’t need a “Hinglish” toggle - you write the way your callers talk, and the agent follows.

  • Tip: if a particular word or brand name must be pronounced a certain way, spell it out phonetically in your prompt. The agent says it the way you wrote it.

Secondary language & mid-call switching

Below the primary language there’s an optional secondary language (it defaults to None). Set one, and the agent will detect when the caller replies in that language and switch to it mid-call - smoothly, without restarting the conversation. A caller who was greeted in Hindi but answers in Tamil gets the rest of the call in Tamil.

This is genuinely useful for mixed lists - a region where some contacts speak Hindi and some speak Punjabi, say - because one agent can handle both instead of you building two.

Don’t add it “just in case”

Here’s the honest tradeoff, and it’s the whole reason this is a separate opt-in setting: every mid-call language switch costs about 200 milliseconds. That’s the agent listening, recognising a different language, and adjusting. A fifth of a second doesn’t sound like much, but on a phone call every pause adds up, and it’s pure waste if the switch never actually happens.

  • Set a secondary language only if your list is genuinely mixed and you expect real callers to answer in a different language than you opened with.
  • If your list is one language, leave it on None. Don’t add a secondary “to be safe” - you’re paying a latency tax on a switch that will never fire.
  • If your list is truly two-way mixed, this is exactly what the feature is for. Use it deliberately, not defensively.

Rule of thumb: one language for the whole list → primary only. Predictably two languages across the list → primary + secondary. Chosen on purpose, never “just in case.”

When you set a secondary language, the voice grid gives you a Preview button for it too - so you can hear your chosen voice in both languages before publishing.

Latency & how it sounds - handled for you

You might expect a page about voice to have knobs for speed, pauses, and turn-taking. There aren’t any, and that’s deliberate. Under the hood the agent is tuned for fast, natural turn-taking: the caller hears a reply quickly after they stop speaking, and the agent slips in short acknowledgements (“one moment”, “let me check that”) so the conversation never feels like it stalled. That’s why the calls sound human rather than like a slow, robotic Q&A - and it’s all managed for you, with nothing to configure.

The only latency lever you control is the one above: whether to add a secondary language. Everything else about how the voice flows is already handled.

Putting it together

A good voice-and-language setup for a Hindi-speaking lead list looks like this:

  • Language: Hindi (primary), None (secondary) - the whole list speaks Hindi.
  • Voice: previewed three warm voices in Hindi, picked Shubh because it sounded the most reassuring.
  • Script: written in natural Hinglish so it matches how the callers actually talk.

That’s it - two decisions, both made with your ears. Before you publish, always test the agent so you hear the voice and language on a real call, exactly as your customer will.

Next: Knowledge base → - give the agent your real documents and FAQs so it answers from your actual content instead of guessing.