# IVR in Any Indian Language with Recorded Prompts (2026) > How to run an IVR in Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada, Malayalam, Odia, Punjabi or any other language using recorded audio prompts: writing the script per language, recording tips, file formats (mp3, wav, m4a, ogg up to 5 MB) and keypad input. Source: https://edesy.in/blog/regional-language-ivr-recorded-prompts-2026 **Short answer:** any language can run on an IVR if the prompts are recorded audio and the caller answers on the keypad. Write the script once, translate it into each language as people actually speak it, record each prompt as a separate short file, upload them, and wire each key to the next step. The menu logic is the same in every language; only the audio changes. This matters in India because built-in text-to-speech covers only some languages well. On Edesy, text-to-speech in the IVR is available for **Hindi and English** today. For Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada, Malayalam, Odia, Punjabi, Assamese, Bhojpuri, Tulu or any dialect your callers speak, recorded prompts are the way to do it, and they work now. ## How a Recorded-Prompt IVR Works A phone menu has two parts: what the caller hears, and what the caller presses. - **What they hear** comes from a **Play** node, which plays an audio file you uploaded. - **What they press** is handled by a **Menu** node, which listens for keypad digits (DTMF) and sends each key down its own branch. Language never touches the second part. A caller in Madurai pressing 1 sends exactly the same signal as a caller in Kolkata pressing 1. So the menu logic, the transfers, the HTTP requests to your system and the call log all work identically. You only swap the audio. A typical flow: 1. **Play**: greeting and menu question in Tamil (one recorded file). 2. **Menu**: keys 1, 2, 3 and 0, with a short timeout. Leave the menu's own text prompt empty so nothing is spoken over your recording. 3. Each key goes to its next step: another **Play** with information, a **Transfer** to a person, a **Voicemail**, or an **HTTP Request** to your system. 4. **Timeout / wrong key**: a **Play** with "we did not get your answer" and a second chance. If you want callers to choose their language, put a language menu first and build one branch per language. Where that menu belongs and how many languages to offer is covered in [multi-language IVR for India](https://edesy.in/blog/multi-language-ivr-india-2026). ## Step 1: Write One Master Script Write the full script in English or Hindi first, as a numbered list of prompts. Every prompt gets an ID, because you will record and upload them one by one: | ID | Purpose | Master text | |---|---|---| | P01 | Greeting + menu | Welcome to City Care Clinic. To book an appointment, press 1. For report timings, press 2. To talk to reception, press 0. | | P02 | Report timings | Reports are ready after 6 p.m. on the day of the test. | | P03 | No input | We did not get your answer. Please try again. | | P04 | Transfer | Connecting you to reception. | Keep prompts short. General rules (option before key, at most four options, no long preamble) are in [how to write an IVR script](https://edesy.in/blog/how-to-write-ivr-scripts-2026). ## Step 2: Adapt, Do Not Translate Give the master script to a native speaker who knows your callers, not to a translation tool. Three things to ask them for: - **Spoken, not written, register.** Formal written Tamil or Bengali sounds like a government announcement. Ask for how a friendly receptionist would say it. - **Keep the English words people use.** "Appointment", "report", "OTP", "order" are usually said in English inside a Telugu or Marathi sentence. Translating them makes the prompt harder to follow. - **Same key numbers in every language.** Key 1 must mean the same thing in every branch, or your staff and your call logs will be confusing. Word order changes between languages. In English you say the option then the key; in most Indian languages the verb, and so "press 1", comes at the end of the sentence. Let the speaker put the key where it sounds natural, as long as each option is one short sentence. ## Step 3: Record Properly You do not need a studio. You do need a quiet room and some discipline. **Room.** A small room with curtains, a bed or sofa, and the door shut. Avoid offices with glass walls and hard floors; the echo is very obvious on a phone line. Turn off fans and AC for the takes. **Speaker.** One person for every prompt in a language. Callers notice when the voice changes between the greeting and the menu. A clear, calm voice at a slightly slower pace than normal conversation works best on phone audio. **Distance and level.** Keep the microphone or phone about a hand's width from the mouth, slightly to the side to avoid breath pops. Do a test recording and check that the loudest word does not distort. **Takes.** Record each prompt two or three times and keep the cleanest. Leave about half a second of silence before and after, then trim it to a short, even gap. **Same session.** Record all prompts for a language in one sitting so the tone and level match. If you must add a prompt later, ask the same speaker and play back an old prompt first to match it. **Numbers and dates.** Recordings cannot change per caller, so avoid putting anything variable in them. If you need a time ("after 6 p.m.") make sure it is genuinely fixed. If you need a quick first draft before booking a speaker, our free [IVR voice generator](https://edesy.in/tools/ivr-voice-generator) can produce a sample audio file you can upload to test the flow. Replace it with a human recording before you go live if the draft does not sound right to a native listener. ## Step 4: Export the Right File The Play node accepts **mp3, wav, m4a or ogg**, up to **5 MB** per file, or a public URL to an audio file. Some practical settings: | Setting | Recommendation | Why | |---|---|---| | Format | mp3 or wav | Most widely supported by recording apps | | Channels | Mono | Phone calls are mono; stereo only doubles the size | | Length | Under 30 seconds per file | Long prompts lose callers | | Loudness | Similar across all files | Callers should not need to change volume mid-call | | Name | `ta_P01_greeting.mp3` | Language code plus prompt ID keeps uploads straight | Phone networks carry narrow-band audio, so studio quality is not audible on the call. A clean, well-levelled mono mp3 is enough. File size is rarely a problem: 5 MB is minutes of mp3 audio, far longer than any prompt should be. ## Step 5: Build and Wire the Flow 1. In the flow builder, add a **Play** node and upload `ta_P01_greeting.mp3`. 2. Add a **Menu** node after it. Leave the text prompt empty. Set the keys (1, 2, 0) with labels such as "Appointment", "Reports", "Reception". The labels are for your call logs and staff, not spoken, so write them in English. 3. Set a timeout. Five to eight seconds suits most callers; older callers may need longer. 4. Connect every key and the timeout path. The builder blocks publishing if a key goes nowhere. 5. For each branch, add the next Play, Transfer, Voicemail or HTTP Request. 6. Use the Loop setting on the Play node if a message should repeat, such as hold information. Then place a real test call on a basic phone as well as a smartphone. Listen for prompts that cut off, uneven volume and gaps that feel too long. ## Step 6: Test With Native Speakers Ask two or three people who speak the language natively, and are not involved in the project, to call and complete a task: "find out when reports are ready", "reach reception". Watch where they hesitate. The usual findings: - The greeting is too long; they start pressing before the options are read. - One option name is a word they do not use. - The timeout is too short for someone holding the phone to their ear. Fix the recording or the timeout, and test again. ## Mixing Recorded Languages With Hindi or English Text-to-Speech A common pattern is Hindi and English with text-to-speech, plus one or two recorded regional branches for your biggest states. Hindi text-to-speech lets you speak names, amounts and dates per caller, so the Hindi and English branches can be more personal. The recorded branches stay general. That is a reasonable trade-off: it gets every caller a menu in their language today, and you can change the Hindi and English wording any time without re-recording. Hindi specifics are in our [Hindi IVR setup guide](https://edesy.in/blog/hindi-ivr-setup-guide-2026). ## Limits to Know - **Recordings are fixed.** No names, amounts or dates per caller in recorded languages. - **Keypad only.** Menus listen for keypad digits, not speech. This is a strength for regional languages: no speech recognition is needed. - **No built-in time-of-day routing yet.** If you want a different message after hours, publish a separate flow or route key 0 to voicemail. ## Build Your Regional-Language IVR See how recorded prompts work on Edesy on the [any-language IVR page](https://edesy.in/ivr/languages/any-language-recorded-prompts). Send us your master script and we can help you plan the recordings: [contact us](https://edesy.in/contact). ## Related Reading - [Multi-language IVR for India](https://edesy.in/blog/multi-language-ivr-india-2026) - [IVR voice prompts: recording vs text-to-speech](https://edesy.in/blog/ivr-voice-prompts-recording-vs-tts-2026) - [What is DTMF? Keypresses and IVR](https://edesy.in/blog/what-is-dtmf-ivr-keypress-guide-2026) - [Hindi IVR: setup, voice options and sample scripts](https://edesy.in/blog/hindi-ivr-setup-guide-2026) ## Frequently Asked Questions ### Which audio formats can I upload for IVR prompts? The Edesy Play node accepts mp3, wav, m4a and ogg files up to 5 MB each, or a public URL to an audio file. A 5 MB limit is generous: a 30-second mp3 prompt is typically well under 1 MB. ### Can the IVR speak a customer's name or amount in Tamil or Bengali? Not with recorded prompts alone, because a recording is fixed. Built-in text-to-speech is available for Hindi and English today. For other languages, keep personal details out of the spoken prompt, or use a fixed recording and send the details by SMS or WhatsApp alongside the call. ### Do callers need to speak their language for the IVR to understand them? No. Callers answer with the keypad, which works the same in every language. That is what makes a recorded-prompt IVR work for any language: only the audio changes, the menu logic stays identical. ### Can I record prompts on a phone? Yes, if you record in a quiet room with soft furnishings, hold the phone about a hand's width from your mouth, and export as mp3 or m4a. A phone recording in a quiet room sounds better than a studio-grade microphone in an echoey office.