AI Voice and Music: Text to Speech, Voice-Over and Songs (ElevenLabs, Suno)

By Ravindra Bagale, trainer · LinkedIn · Last updated: 19 October 2026

Generative AI course · Chapter 27 · Estimated reading time: 13 minutes

1. Suppose a shop needs a voice-over by tomorrow

Suppose we run Sahyadri Agro Tools, a small farm-supply shop in Satara. We made a 40-second phone video showing how to set up our drip irrigation kit. We want a clear voice-over, a short jingle for the shop, and later an intro for a weekly farming podcast.

We copy the kit's instruction leaflet into a text-to-speech tool and press Generate. The result reads out "Kit No. DK-12, Rs 499, 16mm pipe x 10m" as a string of letters. It never pauses. Nobody would listen to it for 40 seconds.

The fix is not a costlier voice. It is a script written for the ear, read by a voice we have the right to use.

2. How AI voice and AI music work

Text to speech (TTS)

1. You type text.
2. The model predicts how a person would say it: the sounds, the speed, the pauses, the rise and fall of the voice.
3. It produces audio in a voice you pick from a library, or a voice you design by describing it ("calm, warm, middle-aged male voice").

Voice cloning

1. You upload a recording of a real voice.
2. The model learns how that voice sounds.
3. It can then say any new text in that voice.

This is powerful, and it is exactly how voice scams work. So cloning is only for your own voice, or a voice whose owner has clearly agreed.

AI music

1. You describe a song: style, mood, instruments, length, and optionally your own lyrics.
2. The model creates the melody, the instruments and the singing together.
3. You get 2 or more versions and pick the best one.

3. Write a script for the ear

Most bad voice-overs come from bad scripts. Read your script aloud once before you generate. If you stumble, the AI will too.

Also:

• Write numbers as words when the tool reads them wrongly.
• Use full stops for pauses. A new sentence gives a natural breath.
• Put the most important line first. People decide in the first few seconds.
• Speaking speed: a calm voice-over is about 150 words a minute. For a 40-second video, aim for about 100 words.

4. Which tools you can use (October 2026)

Prices, credits and features change often. Check the plan page before you plan a paid project, and read the terms for your plan.

What is commercial use? Anything that earns money or promotes a business: an ad, a shop jingle, a client video, a monetised YouTube channel. For these, generate the audio while you are on a paid plan. Both companies say that audio made on the free plan does not become commercial later, even if you upgrade.

5. Worked example 1: an explainer voice-over

Script (rewritten for the ear):

Hello! Welcome to Sahyadri Agro Tools. In the next forty seconds, we will
show you how to set up our drip irrigation kit for a small kitchen garden.
First, connect the main pipe to your tap with the blue connector. Second, lay
the thin drip pipes along each row of plants. Third, push one dripper next to
each plant, about two inches from the stem. Turn on the tap slowly and check
that every dripper gives a steady drop. That is it. Your plants now get water
at the roots, and you save water every day. For help, call the number on the box.

Check the length and the credits first. We counted the script with 3 lines of Python:

t = open("script.txt").read().strip()
words = len(t.split())
print(f"characters: {len(t)}  words: {words}  at 150 words/min: {words/150*60:.0f} seconds")

Real output (we ran this on 19 October 2026):

characters: 541  words: 104  at 150 words/min: 42 seconds

So the script fits a 40-second video, and at about 1 credit per character, one take costs roughly 541 credits. The free 10,000 credits allow many retakes for practice.

Steps:

1. Open the text-to-speech tool and paste the script.
2. Pick one calm voice from the library. Choose a voice that suits Indian English, if the library offers one.
3. Generate. Listen with earphones, from start to end.
4. Fix only the lines that sound wrong: rewrite a number as words, or split a long sentence.
5. Download the audio and add it to your video in a phone editor.

Expected output (yours will be similar, not the same): a clear, calm 40-second voice-over with natural pauses at each full stop. The shop name "Sahyadri" may sound slightly off; if it does, try another voice or spell it the way it sounds (for example "Sah-yaa-dri").

Hindi or Marathi voice-over? ElevenLabs supports Hindi and many other languages. Check its current language list for Marathi. Paste the script in the language you want, and listen to every line, because names and local words are where AI voices slip most.

Remember the plan: this is a shop video, so it is commercial. Practise on the free plan, then generate the final audio on a paid plan, or record it in your own voice.

6. Worked example 2: a jingle for the shop

A jingle is a short, catchy song with the shop name in it. It is the best place to use AI music, and also the best place to make a licensing mistake.

Step 1: write the lyrics yourself. Short lines, the shop name twice, one clear promise:

[Verse]
From the seed to the harvest day,
Sahyadri Agro shows the way.
[Chorus]
Sahyadri Agro, tools that grow,
Satara's farm shop, now you know!

Step 2: describe the style:

Upbeat folk-pop jingle, 20 seconds, dholki and harmonium, cheerful male and
female voices singing together, clear words, ends on the shop name.

Step 3: generate in Suno (Custom mode, so you can paste your own lyrics), listen to both versions, and regenerate until the words are clear.

Expected output (yours will be similar, not the same): 2 short song versions. One usually has clearer words than the other. AI singers sometimes blur or repeat a line; regenerate, or shorten the lyrics.

The licence check: the jingle will play in an ad or a shop video. That is commercial use, so the free Suno plan is not enough. Make the final version while subscribed to a paid plan. The lyrics are yours either way, because you wrote them.

Do not ask for "a song in the style of" a famous singer, or reuse lyrics from a film song. Use your own words and describe the style in general terms (folk-pop, dholki, harmonium).

7. Worked example 3: a podcast intro

Suppose we start a free weekly podcast called Shetkari Sathi Talks for farmers. We want a 12-second intro: 4 seconds of music, then a voice, then the music fades.

Steps:

1. Make a short instrumental in Suno or with ElevenLabs music: "warm acoustic guitar and flute, morning feel, 12 seconds, no vocals".
2. Write a one-line intro for the ear: "This is Shetkari Sathi Talks. One useful farming idea, every Sunday."
3. Record it in your own voice (best), or generate it with a library voice.
4. In a free audio or video editor, place the music first, start the voice at 4 seconds, and lower the music under the voice.

Expected output (yours will be similar, not the same): a 12-second intro you can reuse in every episode.

Free or paid? If the podcast is non-commercial and you follow the attribution rules (for example "elevenlabs.io" or "Eleven Music" for ElevenLabs free-plan audio), the free plans can work. The moment you add sponsors or ads, it is commercial, and you need audio made on a paid plan.

8. Consent and labels for voices

• Clone only your own voice, or a voice whose owner has given clear, written permission for this exact use.
• Never clone a family member, a boss, a celebrity or a politician for fun. A cloned voice saying something they never said is a deepfake. It can be used for scams, and it can hurt real people.
• Tell people when a voice is AI. A line such as "Voice-over generated with AI" in the description is enough for most videos.
• On YouTube, turn on the "altered or synthetic content" disclosure if the AI voice could make viewers believe a real person said something.
• Keep your raw voice recordings private. A few clear minutes of your voice online are enough material for someone else to clone it.

To learn how cloned voices are used in scams, and how to check a suspicious voice note, see the chapter Deepfakes Explained later in this course.

9. Beginner mistakes and quick fixes

10. Try at home

Recap

Write one 30-second script for the ear today and read it aloud before any AI reads it. A good script makes even a free voice sound professional.

Previous chapter: Image-to-Video Explained: Direct a Short AI Clip from One Photo

Next chapter: AI for YouTube and Instagram Reels: From Script to Captions, Plus a 30-Day Plan

Ravindra Bagale, trainer: linkedin.com/in/ravindra-bagale. The shop, podcast and people on this page are made up for learning. Plans, credits and licence rules were checked on ElevenLabs' pricing and help pages and Suno's help centre on 19 October 2026 and can change. The Python output was produced on 19 October 2026. AI audio changes every time, so your output will be similar, not the same.