How to Make an AI Voiceover: Script, Pronunciation Fixes and File Delivery

The four tools at a glance

In shortThree tools turn text into speech; Auphonic only finishes audio that already exists. The two free-route columns decide whether a free plan can produce the file you hand over.

Tool	Step it covers	Free: download	Free: publish commercially	First paid plan	Usage counted in
TTSMaker	Text to speech in the browser, stock voices	Yes, MP3; save it before the short-lived history expires	Yes, under its terms, if you hold the rights to the text	Lite; the total is confirmed at checkout	Characters: 20,000 a week on Free
ElevenLabs	Text to speech in 74 languages, cloning and dubbing	Yes, for noncommercial use	No: noncommercial, attribution required, and a later upgrade does not change it	Starter, USD 6 billed monthly (USD 5 billed yearly)	Credits: 1 per character on Multilingual v2
Murf	Text to speech in a studio with images, video and music	No: previews and share links only	No	Creator, USD 29 billed monthly (USD 19 billed yearly)	Minutes of generated speech: 10 one-time on Free
Auphonic	Finishing: leveling, noise reduction, loudness targets	Yes	Commercial terms are not covered in the sources checked; free productions carry an Auphonic jingle	Auphonic S, USD 13 billed monthly (USD 11 billed yearly)	Hours of processed audio: 2 a month on Free

Plan names and amounts from each tool’s profile, which shows the date its plan terms were checked. TTSMaker’s paid totals could not be confirmed with a single billing period and currency, so none is shown.

Visit TTSMaker (opens in a new tab) Visit ElevenLabs (opens in a new tab) Visit Murf (opens in a new tab) Visit Auphonic (opens in a new tab)

Free plans: what the file can be used for

In shortA free plan that downloads is not the same as one whose audio you may publish. Check both before the first take.

ElevenLabs fixes rights at the moment of generation, so render the final take on the plan you will publish under and keep Free for auditions that stay unpublished. Murf’s free trial is for auditioning Studio voices, and its unused free minutes are added to the paid plan after an upgrade.

Sources: TTSMaker free web converter, TTSMaker terms of service, ElevenLabs publishing rights, Murf downloads on the free trial, Murf pricing.

Step 1: Agree the file before you choose a voice

In shortAsk for format, channels, sample rate, maximum length and any loudness target first. Then pick the route that can hand over that file on the plan you can use.

If the recipient gives no specification, agree one before rendering. A web preview is not a downloadable file, and renaming an MP3 to .wav does not change its encoding.

Route	Files listed	Plan needed	Check before you start
TTSMaker	MP3 downloads from the web converter; API results arrive as a temporary URL valid for 24 hours	Free for web downloads; API from PRO Mini, not on Free or Lite	WAV or other formats are not covered in the sources checked
ElevenLabs	44.1kHz PCM output via the API and 192kbps audio, both listed on Pro	Pro, USD 99 billed monthly, for those two	File formats on lower plans are not covered in the sources checked
Murf	MP3, WAV, FLAC, a-law or u-law up to 48 kHz; video as MP4 or MOV with optional subtitles	Creator or higher: Free cannot download	Music tracks cannot be exported on their own
Auphonic	Many audio formats plus video, chapters and transcripts, per its production help	Free for finishing, with a jingle; S and up without it	Speech recognition, shownotes and chapters are paid

Sources: TTSMaker API documentation, ElevenLabs pricing, Murf voice export formats, Murf video export, Auphonic production settings.

Step 2: Keep the approved script apart from the spoken script

In shortThe approved script fixes meaning and spelling. The spoken copy says the same thing in words a voice reads correctly.

Keep two text files. Changing “14:30” to “two thirty p.m.” keeps the meaning; changing the appointment time does not. When a name has no agreed pronunciation, ask the script owner instead of guessing. The table shows the kind of rewrites to try on a short fictional announcement.

These are wording suggestions, not guaranteed pronunciations. Enter plain text first: tags or pause syntax written for one product can be read aloud or dropped by another.

Step 3: Check the limits for your script length

In shortEach tool caps one piece of text differently. A long script may have to be split before it is converted.

Tool	Limit on one piece of text	What it means for a long script
TTSMaker	The Free comparison lists 500 characters per conversion; voices in the live converter show their own limits, such as 1,000 or 2,000. Lite takes up to 10,000, PRO Mini 20,000	Split the script on Free and keep the parts numbered; check the limit on the voice you picked
ElevenLabs	Studio projects for long-form audio: 3 on Free, 20 on Starter	A per-request character limit is not covered in the sources checked
Murf	15,000 words per script upload; 1,000 characters per block in the Canva app, with 50 audio blocks per design	Upload long scripts in parts; in Canva, Free Trial users can only preview
Auphonic	Multitrack productions under 20 minutes without paid credits	A single narration file is not affected

Sources: TTSMaker free web converter, TTSMaker plan comparison, ElevenLabs pricing, Murf import limits, Auphonic pricing.

Step 4: Fix pronunciation, pauses and pacing

In shortStart with respelling in the spoken copy, which works everywhere. Use a tool’s own controls only after checking that your plan and voice offer them.

• TTSMaker lists speed, volume and pitch controls and pause insertion. Some advanced controls need a paid plan and depend on the voice; no pronunciation dictionary is documented, so fixes stay as respellings.
• Murf adjusts pitch, speed, pauses and pronunciation block by block, with custom IPA, an alternative spelling or a locale change for a word, applied to one block or the whole project. Changing the voice, pitch, speed or pronunciation of unchanged text uses no voice generation time.
• ElevenLabs takes pronunciation dictionaries with phoneme or alias rules. Phoneme rules work only on the eleven_v4, eleven_flash_v2 and eleven_v3 models, so on other models use alias rules, which swap a word for a written replacement.

Keep the decision for each difficult word in one list with the written form, the approved spelling and the respelling that worked, so the next script starts from it. The ElevenLabs vs Murf comparison covers the two studios side by side.

Sources: TTSMaker free web converter, Murf pronunciation help, Murf voice generation time, ElevenLabs pronunciation dictionaries.

Step 5: Audition one sentence, then correct one thing at a time

In shortRender the hardest sentence before the whole script, and change one variable per take so you know what fixed it.

Pick the sentence with the most important number or correction. Listen once without reading, write down what you heard, then compare it with the approved script; this catches the word your eyes would otherwise supply.

Set a small take budget before you start; retakes are charged differently in each tool, and the cost section below counts one.

Sources: ElevenLabs credits, Murf voice generation time.

Step 6: Finish the sound only after the words are approved

In shortLeveling and loudness cannot repair a wrong date. Approve the words, keep an untouched original, then make a separate finished file.

If the recipient names a loudness target, use it and record the result. See the Auphonic profile for plans, limits and sources. levels speech, reduces noise and reverb, cuts fillers and silences, and produces audio to targets such as integrated loudness, true peak and MaxLRA, for example -16 LUFS for podcasts or EBU R128. Re-running a production with different settings costs nothing while the input file stays the same; a new input file or a new production is charged again, for at least 3 minutes each time.

Sources: Auphonic features, Auphonic pricing FAQ, Auphonic pricing.

Step 7: Deliver a version someone else can verify

In shortHand over the approved file, both scripts and a short record. Anyone can then trace what was generated, on which plan, under which terms.

Keep the delivered file apart from auditions and rejected takes, and name it so the extension matches the encoding, such as q7-announcement-v03-approved.mp3. Supply a WAV master only when the route really exported one; converting an MP3 to WAV cannot restore what was discarded.

Delivery record: fill from actual evidence
Approved audio filename and version:
Approved script / spoken script filenames:
Product, plan and generation date:
Voice, model and settings:
Format, channels, sample rate, length:
Loudness target and measured result:
Usage terms URL; attribution or jingle:
Known limitations and recipient approval:

If the voiceover goes on a video, Murf’s Studio places it on a timeline with images, video and music and exports MP4 or MOV on paid plans; otherwise pass the approved script and matching audio to the video captions workflow, because a text change after approval moves caption timing.

Sources: ElevenLabs publishing rights, TTSMaker free web converter, Murf video export.

A 1,000-word script, auditioned, rendered and fixed twice

In shortThe same job counted in each tool’s unit: one 200-character audition, one full render, two rewritten paragraphs and one voice change after the client listens. Credits from different tools are never compared with each other.

Murf estimates about 6 minutes of speech per 1,000 English words, and ElevenLabs about 1,000 credits, one per character, per minute of speech, so the script is taken as about 6,000 characters. Each rewritten paragraph is about 600 characters, or 0.6 minutes. Take counts are assumptions; replace them with yours.

Tool	How a take is charged	The arithmetic	What that means for the plan
ElevenLabs	Every regeneration, at 1 credit per character on Multilingual v2	200 + 6,000 + 2 × 600 + 6,000 for the voice change = 13,400 credits	Above Free’s 10,000, which is noncommercial anyway; Starter’s 30,000 cover two such scripts a month
Murf	Minutes of generated speech; a text edit regenerates its sub-block, a voice change on unchanged text is free	0.2 + 6 + 2 × 0.6 + 0 = 7.4 minutes, if each fix stays inside one sub-block	Free’s 10 one-time minutes cover it but cannot download; Creator’s 120 a month on monthly billing cover about 16 (120 / 7.4)
TTSMaker	Characters converted, if the voice charges the full text each time	200 + 6,000 + 1,200 + 6,000 = 13,400 characters	Inside Free’s 20,000 a week, but the 6,000-character render has to be split: 3 parts at 2,000 or 12 at 500. Lite takes it in one
Auphonic	Length of each production, at least 3 minutes; re-runs on the same input are free	Finishing once: 6 minutes. If both fixes arrive after finishing, two new inputs: 6 + 6 + 6 = 18 minutes	Inside Free’s 120 minutes a month, with a jingle; a 30-second audition finished on its own still counts 3

Two rules follow from the table. Ask for the client’s voice decision before the full render, because a voice change after it costs a full render on ElevenLabs, and on TTSMaker voices that charge the full text, but nothing on Murf. Finish in Auphonic only after the words are approved, so a fix does not start a new production.

Sources: ElevenLabs pricing, ElevenLabs credits, Murf pricing, Murf voice generation time, TTSMaker plan comparison, Auphonic pricing FAQ.

In shortFourteen capabilities marked from the four profiles. A dash means the sources checked do not cover it, not that the tool lacks it.

Capability	TTSMaker	ElevenLabs	Murf	Auphonic
Making the speech
Text to speech with stock voices	Offeredlanguage and voice choice	Offered74 languages	Offered200+ voices	Not listedfinishing only
Pronunciation control	Not listedno dictionary documented	Offeredphoneme or alias rules	OfferedIPA, spelling, locale	Not listednot covered in the sources checked
Pauses, speed and pitch	Offeredsome controls paid	Not listednot covered in the sources checked	Offeredblock by block	Not listednot covered in the sources checked
Voice change without a new charge	Not listednot covered in the sources checked	Not listedevery regeneration charged	Offeredon unchanged text	Not listednot covered in the sources checked
Voice cloning	Not listednot covered in the sources checked	Offeredinstant from Starter	PartlyEnterprise team	Not listednot covered in the sources checked
Video, images and music on a timeline	Not listednot covered in the sources checked	Not listednot covered in the sources checked	OfferedStudio	Not listednot covered in the sources checked
Finishing
Loudness targets	Not listednot covered in the sources checked	Not listednot covered in the sources checked	Not listednot covered in the sources checked	OfferedLUFS, true peak, MaxLRA
Noise and reverb reduction	Not listednot covered in the sources checked	PartlyVoice Isolator, recordings	Not listednot covered in the sources checked	Offeredevery plan
Filler and silence cutting	Not listednot covered in the sources checked	Not listednot covered in the sources checked	Not listednot covered in the sources checked	Offeredevery plan
Getting the file out
Download on the free plan	OfferedMP3	Offerednoncommercial	Not listedpreviews only	Partlywith a jingle
Commercial use	OfferedFree, under its terms	Partlyfrom Starter	Partlyfrom Creator	Not listednot covered in the sources checked
WAV export	Not listednot covered in the sources checked	PartlyPCM via API on Pro	Offeredpaid plans, up to 48 kHz	Not listednot covered in the sources checked
Video export with subtitles	Not listednot covered in the sources checked	Not listednot covered in the sources checked	OfferedMP4 or MOV	Partlyvideo output listed
API	Partlyfrom PRO Mini	OfferedElevenLabs API	Offeredseparate API billing	OfferedREST API

offered partly or with limits not listed on the profile.

Cells come from each tool’s profile on this site, which cites the vendor pages and the date they were checked.

Sources: TTSMaker plan comparison, ElevenLabs pricing, Murf pricing, Auphonic features.

Common questions

What this guide does not judge

The pages that rank for this search lead with realistic voices, emotional range, the size of the voice library and how fast audio is generated. None of the vendor pages behind this guide document those in a way we can check, so no tool is ranked on them; the same goes for multi-speaker scenes and ease of use.

Clipchamp, Epidemic Sound, Voicebooking, MiniMax and Google AI Studio appear in the same results. They are left out until each has a profile with checked plan terms, and the Canva, Adobe Firefly and Speechify profiles do not yet cover their voiceover plan terms. For which free plans let you keep and publish an output in other tasks, see the free AI tools guide; more narration tools sit in the audio category.

How this guide was made

Plans, units, limits, controls and file formats come from the TTSMaker, ElevenLabs, Murf and Auphonic profiles on this site, which cite the vendors’ pricing, help and terms pages and carry their own check dates; pronunciation controls come from the ElevenLabs and Murf help pages listed above. Sections follow the topics the pages ranking for this search cover, limited to what the vendors document. The practice script and diagrams are editorial material; no audio was generated for this guide and no rating is given. See our editorial policy.