How to choose a voice for a faceless YouTube channel and keep it the same in every video

On a faceless channel nobody sees you, so the voice is the presenter. It is the one thing that should be the same in every video. Two things go wrong. You choose a voice from a ten-word preview and it does not suit your scripts. Or you find a good one, and three months later you cannot remember which voice it was or what the speed was set to. This guide covers both: how to audition voices fairly, and how to write the settings down so that episode fifty sounds like episode one.

Audition every voice with the same script

A preview clip tells you how a voice reads someone else's sentence. You need to know how it reads yours. Write one audition script of about thirty seconds and use it, unchanged, on every voice you try. If the script changes, you are comparing scripts and not voices. Build it from the lines your channel says in every episode:

• Your opening line and channel name. The voice will say them more often than anything else.
• A question. You hear what the voice does at a question mark.
• A number, written as a word. “पाँच मिनट”, not “5 मिनट”.
• An English word or two, such as channel or subscribe, if your Hindi scripts use them.
• Your closing line.

For an English channel, choose English in the Language list and use the same idea:

Shortlist three, then choose one

Without logging in you can audition the free Microsoft voices, 500 characters at a time and 2,000 a day. That is six runs of a script this size. For Hindi there are two, Madhur (male) and Swara (female).

A free account (log in with Telegram) opens the other three engines in Studio. For Hindi the catalogue has 101 ElevenLabs voices and 30 Gemini AI voices today, and the voice picker filters by language and by male or female. One audition of this script costs about 310 credits on an ElevenLabs or Cappy voice and about 620 on a Gemini AI voice, so the 5,000 welcome credits on a new account cover around 16 auditions at the first rate, or 8 at the second.

Male or female, calm or energetic: that is your call, and no rule settles it. Pick three voices, convert the audition script with each, and play the three files one after another. We cannot hear the voices for you, so the test that counts is your script in your ears.

What the voice costs per episode

A channel voice is a running cost, so work it out before you get attached to one. An eight-minute video is about 7,200 characters of script, at roughly 900 characters a minute.

Credits bought with a plan are valid for 30 days, so a premium voice means buying credits in every month you publish. The 1,000,000-credit plan covers about 138 such episodes on an ElevenLabs or Cappy voice, or about 69 on Gemini AI; the pricing page has the current prices. If you do not want to carry that cost for a year, choose a free voice now. Switching later gives up the consistency you are building.

Write a voice card

Once you have chosen, write down every setting in the last column of the table. On the website they are in Studio under Voice settings; in the Telegram bot they are under Voice Settings.

• Microsoft Edge: Speed runs from −50% to +100% and Pitch from −50 Hz to +50 Hz.
• ElevenLabs: do not skip the Model. As the bot's own settings screen puts it, the model decides how it sounds and the voice decides who speaks.
• Gemini AI: copy the Speaking style text word for word. The box holds up to 200 characters.

Keep the card where you keep your scripts. Write down the settings you did not change as well: a new account starts at Speed 0%, Pitch 0 Hz, Stability 0.50, Similarity 0.75, Style exaggeration 0 and Speaker boost on, and “I never touched that” is hard to be sure of a year later.

What TTS Maker remembers, and what it does not

The card matters because the tool holds only your current settings, with no record of what you used before.

• Studio saves your settings to your account. The engine, the voice and every slider stay as you left them, and the Telegram bot uses the same ones.
• The quick converter on this page saves nothing. Reload the page and it is back to the default voice, with Speed and Pitch at 0.
• Changing the engine changes the voice. Tap another engine in Studio and the voice switches to that engine's first voice. Come back to Microsoft Edge and the voice is Madhur, whichever one you had before. The sliders are kept. The voice is not.
• One account holds one set of settings. If you run two channels with two voices, keep a card for each.
• History does not list your settings. On the website and in the bot it shows the engine, the character count and the cost of each audio, not the voice, the speed or the pitch.
• The bot has a Reset button. In Voice Settings, one tap puts the model, the sliders, the Gemini style, the speed and the pitch back to their defaults. The voice stays, the tuning goes.

One record does exist. Audio the bot makes from your text is captioned with the voice name, and if you turn on Send to my Telegram too in Studio, audio made on the website arrives in Telegram with the voice name as well. Speed and pitch are not in the caption.

Keep a reference clip

Download the audition file made with your final settings and keep it with the card. That file is what your channel sounds like. Do not count on History for it: audio is removed after 30 days, so on a weekly channel the first episode's audio is gone before you make the sixth.

Before each new episode:

1. Open Studio and compare the engine, the voice and the settings with the card.
2. Convert the audition script and play it next to the reference clip.
3. If the two match, convert the episode. If they do not, fix the settings first.

The same settings can still sound a little different

Matching settings give you the same voice. On the ElevenLabs engine they do not always give you an identical performance. ElevenLabs says in its own documentation that generation is nondeterministic: the same voice, settings and model give slightly different output each time, and the Stability slider controls how wide that variation is. A lower Stability gives a more varied, more performed read. A higher one is steadier, and set too high it can sound monotonous.

For a channel voice, that is a reason to avoid a low Stability value. The same page recommends leaving Style exaggeration at 0 in general, and notes that raising it can make the model less stable. If one section comes out with a different energy from the rest, convert that section again and leave the settings alone.

What an AI voice cannot give your channel

• It is not yours alone. Every voice here is a catalogue voice, and any other creator can choose the same one. TTS Maker has no feature for cloning a voice or making a private one. If your channel needs a voice nobody else has, record your own.
• A library voice can be withdrawn. TTS Maker is an independent service. It is not affiliated with ElevenLabs or any other voice provider and does not control their catalogues. ElevenLabs' help centre explains that the owner of a Voice Library voice can stop sharing it, after a notice period of between 30 days and 2 years. We cannot promise that a particular voice will always be there, which is why the card has a second choice on it.
• It does not settle YouTube's rules. Monetisation and AI disclosure are not covered here. The Hindi voiceover guide quotes YouTube's own pages on both, and goes deeper into writing the script.

Making short clips for the same channel? Keep the same voice and add a second line to the card for them. The Reels and Shorts guide explains why short video usually wants a faster speed.