Voice Over File Delivery: Formats and Naming
← Guides

Editing

Voice Over File Delivery: Formats and Naming

Updated October 16, 2026·16 min read
Collage of audio file icons with tidy file names, a waveform strip and a sample rate dial

The short answer: when a voice-over brief says nothing, deliver an uncompressed WAV at 48 kHz and 24-bit, mono, unlimited and unnormalised, named Client_Project_Section_Version, and say in your note that you can re-export to any other spec. When there is a published spec, it wins: audiobooks for ACX go as 192 kbps constant-bit-rate MP3 at 44.1 kHz with RMS between -23 and -18 dB, and podcast episodes go as AAC or MP3 at around -16 LUFS with true peaks at or below -1 dBTP.

A surprising number of voice-over jobs never specify a file format, sample rate or naming convention at all. The client simply expects a file that opens, sounds right, and is named something they can find again in a folder of forty other deliverables. Getting this wrong rarely costs you the job, but it costs you something quieter: the impression that you are new to this. Getting it right, consistently and without being asked, is one of the cheapest ways to look like a working professional rather than a hobbyist with a nice microphone. Podcasters face the same problem from the other side: the platforms do publish specs, but they are spread across half a dozen help pages written in different units.

This guide puts the whole delivery job in one place. It covers the three file formats you will actually use, what sample rate and bit depth to pick for each kind of work, mono versus stereo, the loudness each destination expects, a naming scheme with real examples, how to lay out a multi-file delivery, the metadata worth filling in, how to send large files, and exactly what to ask when the brief is blank. Every number is taken from the publisher's own documentation, linked at the end.

WAV vs MP3 vs FLAC: The Three Formats That Matter

WAV is the default professional delivery format. A standard WAV holds uncompressed PCM audio, so what the client receives is exactly what left your session, sample for sample. Nothing is thrown away, nothing is added, and any editor, video tool or broadcast system will open it. The cost is size: one minute of mono audio at 48 kHz and 24-bit is 48,000 samples a second times 3 bytes, about 8.6 MB a minute or roughly 518 MB an hour. The broadcast industry's variant, Broadcast Wave (BWF, defined in EBU Tech 3285), is still a .wav file; it simply adds a block of metadata, covered later in this guide.

MP3 is lossy. The encoder discards detail it predicts you will not hear, which is why a 128 kbps MP3 is about a tenth the size of the same audio as a 16-bit stereo WAV. That makes it the right format for things people listen to rather than things people edit: podcast episodes, audiobooks on ACX (which requires MP3), casting and audition submissions, and review copies a client will play on a phone. It is the wrong format for a master that someone will cut, mix or re-encode, because every further lossy encode compounds the damage. If a client asks for MP3 as the final delivery, give them MP3; just do not choose it for them.

FLAC is lossless compression: it shrinks the file but decodes back to an identical copy of the original audio, the way a zip file does for documents. The Library of Congress describes it as an open format designed for lossless compression of PCM audio. Apple Podcasts Connect accepts FLAC alongside WAV and MP3 for audio uploaded directly for subscriptions, and some archives and music services prefer it. Most voice-over clients never ask for FLAC, and plenty of video editors will not import it, so treat it as a format to supply when asked rather than a default. AAC, usually in an .m4a or .mp4 file, is the other lossy format worth knowing: Apple recommends it over MP3 for podcast feeds and YouTube expects it inside video.

Never convert MP3 back to WAV

Turning an MP3 into a WAV makes a large file of lossy audio. Nothing comes back. If you only have an MP3 of a take, say so; do not dress it up as a master. Always export each format from the original session, not from another export.

Sample Rate and Bit Depth, by Kind of Job

Sample rate is how many times a second the audio is measured. A system can only capture frequencies up to half its sample rate, so 44.1 kHz reaches 22.05 kHz and 48 kHz reaches 24 kHz, both above the roughly 20 kHz limit of human hearing. For voice, neither sounds better than the other. The choice is about where the file is going. 44.1 kHz is the CD and music standard, and it is what ACX requires for audiobooks. 48 kHz is the video and broadcast standard: YouTube's upload recommendations specify 48 kHz audio, and Apple's HLS encoding guidance for video uses 48 kHz AAC. A 44.1 kHz voice track dropped into a 48 kHz video timeline forces a sample-rate conversion on the editor's side, one more step that can go wrong, and in some setups a file read at the wrong rate plays slightly fast or slow and drifts out of sync with picture.

Bit depth sets the dynamic range of each sample, roughly 6 dB per bit: about 96 dB for 16-bit and about 144 dB for 24-bit in theory. 24-bit gives you room to record with comfortable headroom and still keep the noise of the format itself far below your room noise. Deliver 24-bit unless asked for 16. When a client does want 16-bit (some older systems and CD-style workflows), reduce it once, at the very last export, with dither switched on in your editor. Some studios and game teams now ask for 32-bit float for raw takes, because a float file cannot clip internally; if you see that on a spec sheet, it is a real request, not a typo.

  • Commercials, promos, corporate, e-learning, explainers, games and anything going to video: WAV, 48 kHz, 24-bit, mono.
  • Audiobooks for ACX: MP3, 44.1 kHz, constant bit rate of 192 kbps or higher. Record and edit at whatever your session uses, then export to this.
  • Podcast episodes for a feed: AAC or MP3, 44.1 or 48 kHz, around 128 kbps for a mono talk show. Keep a 48 kHz, 24-bit WAV master of every episode.
  • Raw takes for an editor, mixer or game audio team: WAV at 48 kHz, 24-bit (or 32-bit float if asked), unprocessed.
  • Casting and audition files: MP3 at 192 to 320 kbps is normal unless the posting says otherwise, because the listener is judging the read, not the file.
Defaults by destination, from each publisher's own page. When the brief names a spec, use the brief.

Mono or Stereo?

One microphone recording one voice makes mono audio. Exporting it as stereo just writes the same signal twice, doubling the file size and adding nothing. So the default for voice-over is mono, and for podcasts a talk show is usually best published in mono too: Apple's own bitrate table lists mono ranges, Spotify accepts mono, and mono plays identically on a single phone speaker. Stereo earns its place when there is something to put in it: music beds, sound design, or a produced mix.

There are three exceptions worth memorising. ACX requires that every file in an audiobook be mono or every file be stereo, and mixing the two causes errors, so pick one at the start of a book. Apple Podcasts Connect rejects single-channel WAV and FLAC files and asks for two identical channels if your source is mono. And video destinations generally expect stereo audio in the final mix, although the voice track you hand the editor can, and usually should, stay mono. One measurement trap follows from all this: a mono file and the same audio as dual-mono stereo can read about 3 LU apart on a loudness meter, depending on how the meter treats mono, so measure the exact file you are sending.

Loudness Targets by Destination

Loudness is the spec most often misunderstood, because different industries use different meters and different numbers, and because a raw voice track for a mixer should not have a loudness target at all. The modern unit is LUFS (Europe) or LKFS (North America), the same measurement under the international standard ITU-R BS.1770: an average of how loud a whole programme feels. True peak, in dBTP, catches the brief overshoots between samples that appear when audio is converted to MP3 or AAC. ACX is the odd one out and still measures RMS in dB, which is why its numbers look different.

  • European broadcast (EBU R 128): programmes normalised to -23 LUFS, with true peaks no higher than -1 dBTP in production. Its supplement for short-form content such as adverts and promos also caps short-term loudness at -18 LUFS.
  • US television (ATSC A/85 and the CALM Act): -24 LKFS is the reference level. The FCC's CALM Act rules apply A/85 to commercials on television; the FCC notes they do not cover radio commercials or ads on internet streaming services.
  • Podcasts (Apple and Spotify): Apple recommends -16 LKFS with a tolerance of plus or minus 1 dB and true peaks no higher than -1 dBFS. Spotify normalises playback to -14 LUFS by default. A file at -16 LUFS and -1 dBTP fits both, as our podcast audio standards guide explains in detail.
  • On-demand streaming in general (AES TD1008): the Audio Engineering Society's recommendation for internet audio sets -18 LUFS as the distribution loudness for speech and mixed speech-and-music content, and the same for interstitials such as ads and trailers.
  • Audiobooks (ACX): RMS between -23 and -18 dB in every file, peaks no higher than -3 dB, noise floor no higher than -60 dB RMS. The details and the usual failure points are in ACX audio requirements.
  • YouTube: no published loudness target. Its help pages give encoding settings only.
  • E-learning and corporate: there is no industry-wide published standard. Ask, and if nobody knows, deliver clean, consistent, unlimited files and let the developer level them.

Raw voice for a mixer gets no loudness target

Broadcast and streaming numbers describe the finished mix. If someone else is mixing your voice with music and effects, send it clean and unlimited with healthy headroom, and leave the final loudness to them. Hitting -16 LUFS on a voice track that will sit under a music bed only squashes it before the mixer has touched it.

For podcasters who deliver their own finished episodes, the Loudness Normalizer measures the exported file and delivers it at -16 or -14 LUFS with true peak under -1 dBTP, so the number in your notes matches the number the platform measures. For the background on why the platforms land where they do, Podcast Loudness Standards walks through it.

A File Naming Convention That Survives a Client's Folder

Name files so they sort and identify correctly with zero context. A bare Take1.wav or VO_final_FINAL2.wav means nothing once it is sitting in a folder alongside twenty other vendors' equally vague files. The scheme that works reads from broadest to most specific: client, project, section, role, version, and optionally date. Each part answers a question the client will otherwise have to ask you.

Broadest to most specific. Every segment answers a question the client would otherwise email you about.

A few rules make any scheme robust. Use underscores between parts and hyphens inside them, never spaces: spaces break some upload tools and turn into %20 in links. Write dates year first, as ISO 8601 does (2026-10-07 or 20261007), so they sort chronologically everywhere. Pad numbers with leading zeros (M01 to M12, L001 to L250) so module 10 does not sort before module 2. Stick to plain letters and numbers. Windows reserves the characters < > : " / \ | ? * and refuses them in file names, and ACX asks for standard US alphabetical and numeric characters only. Never use the word final. There will be a revision, and "final_v2" is how folders turn into archaeology. Use v1, v2, v3 instead.

Here is how the same scheme adapts to the kinds of job voice actors and podcasters actually deliver:

  • Single commercial with alternates: Brightside_Spring30_Main_v1.wav, Brightside_Spring30_AltA-warmer_v1.wav, Brightside_Spring15_Cutdown_v1.wav.
  • E-learning course: Acme_Safety_M01_Narrator_v1.wav through Acme_Safety_M12_Narrator_v1.wav, then Acme_Safety_M04_Narrator_v2.wav for the revised module only.
  • Video game or line-by-line script: use the client's line IDs exactly as they appear in the script, for example GuardCaptain_L0142_take2.wav. Game teams import by file name, so their scheme is not a suggestion.
  • Audiobook for ACX: one file per chapter, plus separate opening and closing credits, for example SilentHarbor_00_OpeningCredits.mp3, SilentHarbor_01_Chapter01.mp3, SilentHarbor_99_ClosingCredits.mp3. Each file must also open with its spoken section header, such as "Chapter One".
  • Podcast episode: ShowName_E047_2026-10-07_master.wav for your archive, and ShowName_E047.mp3 for the upload. Podcasters receiving guest tracks can ask for Guest-Surname_ShowName_E047.wav so every remote recording arrives already labelled.
A voice actor at a home studio desk renaming a row of WAV files in a folder window on her laptop, headphones around her neck and a microphone on an arm beside the screen
Two minutes renaming files before you send them saves the client twenty minutes of guessing.

If the client sends a naming scheme, a line list with file names, or a spreadsheet of takes, use theirs exactly, character for character, including their capitalisation, even where it breaks the rules above. Their asset system, their editing templates and their invoicing were built around it, and a file that does not match will either fail to import or be renamed by hand by someone who then remembers that you made them do it. Your own scheme is the fallback for the many jobs that never specify one, and the more consistently you use it, the faster your own archive becomes to search when a client comes back a year later asking for the module 4 pickup.

Folder and Zip Structure for Multi-File Jobs

A single commercial needs no structure: one correctly named file and a short note. Anything with more than a handful of files, alternates or formats needs a folder that explains itself. Number the subfolders so they sort in the order you intend, put the finished files where the client will look first, and add a plain text README at the top level with the spec and anything they need to know. Then zip the whole folder, because a zip keeps files together through email links, downloads and cloud previews that otherwise flatten or reorder them.

Numbered folders sort themselves. The README answers the questions before they are asked.

Do not expect the zip to make anything smaller. Audio does not compress much as generic data, and MP3 and AAC are already compressed, so the zip of a WAV delivery is nearly the size of the WAVs. Its job is organisation. Before you send it, unzip your own zip into a fresh folder and play one file from each subfolder. That single check catches the broken archive, the file that never finished exporting, and the 44.1 kHz file that slipped into a 48 kHz batch.

Metadata: The Fields Worth Filling In

Metadata travels inside the file, so it survives even when someone renames it. For WAV deliveries, Broadcast Wave adds a bext block defined in EBU Tech 3285: a Description of up to 256 characters, an Originator, and an OriginationDate written YYYY-MM-DD, among other fields. Version 2 of the spec, published in 2011, added loudness fields based on EBU R 128. Most editors fill some of these automatically; it is worth putting the project, section and your name in the description so the file identifies itself in a broadcaster's library.

MP3 and AAC files carry tags instead: ID3 for MP3, and similar fields in the MP4 container used by AAC. For podcast episodes the hosting platform usually writes these, but Apple notes that loudness information can be carried in ID3 tags or an MP4 header for its Sound Check feature, and Spotify asks that embedded artwork, ID3 tags and other metadata in uploaded MP3s be kept as small as possible. For audiobooks, ACX takes title, author and narrator from your project and expects the spoken opening credits to match. For commercial voice-over, a clear file name matters far more than tags, because many editing tools never display them.

How to Send the Files

Email works for a single short file and almost nothing else. Personal Gmail accounts have a 25 MB attachment limit, and above it Gmail swaps the attachment for a Google Drive link automatically; one minute of 48 kHz, 24-bit mono WAV is already about 8.6 MB, so a three-minute explainer will not fit. For anything larger, send one link to a cloud folder or a file transfer service, set an expiry date the client can live with, and say in your message how long the link will work. Many agencies and production companies have their own upload portal or shared folder; when they do, use it even if your own method is faster, because that is where their workflow starts.

Whatever the method, keep your own copy of every delivery, session and export. Clients come back months later for a pickup or a re-export, and the voice actor who can resend the exact files the same afternoon gets the next job. Our guide to backing up voice-over business files covers how to keep that archive without paying for storage you do not need.

The Five-Minute Pre-Delivery Checklist

  1. Confirm the spec against the brief: format, sample rate, bit depth, channels and, if given, loudness. Check the exported file's properties, not your session settings.
  2. Listen to the first and last three seconds of every file for clicks, chopped breaths and cut-off words. For ACX, check there are 1 to 5 seconds of room tone at the start and end, never more than 5.
  3. Check levels match across the batch. A module recorded on a different day can sit noticeably louder or quieter than the rest.
  4. Play each file fresh in a separate player, not your editor, which can mask problems a plain player will not.
  5. Check the names against the scheme, the line list or the chapter list, and count the files against the script.
  6. Unzip your own zip and test it, then send one link with a short note.

Most of what goes wrong at this stage starts earlier. Background hiss, hum and fan noise are a recording problem, and Studio Rescue removes them on a voice track; it cannot fix echo, which has to be solved in the room. For podcasters, How to Remove Background Noise from Podcast Audio covers that end to end. Breaths, false starts, retakes and dead air are an editing job, which Auto Clean Up handles by transcribing the recording so you edit the audio by editing the words; it quietens breaths in place rather than cutting them, and its keep-in-sync mode for multi-track episodes makes no cuts at all. If you are not sure whether your room itself is the problem, the free Home Studio Analysis measures it.

↗ Try the tool

Loudness Normalizer

Delivering a podcast or narration file to a platform target? Loudness Normalizer measures the file and returns it at -16 or -14 LUFS with true peak under -1 dBTP, ready to upload.

Open Loudness Normalizer →

What to Do When a Job Doesn't Specify Anything

Ask, but ask efficiently. One short message with every question in it gets answered faster than a back-and-forth, and it shows you know what matters. Something like this works for almost any job:

A delivery question you can copy

Before I export, a quick check on delivery: is WAV, 48 kHz, 24-bit, mono right for you, or do you need another format? Should I deliver the voice raw for your mixer, or processed and levelled? Do you have a naming scheme or line list I should follow? And is there a folder or portal you would like the files uploaded to?

When a client genuinely does not know, or does not answer in time, default to WAV, 48 kHz, 24-bit, mono, unprocessed beyond your normal clean edit, with clearly named files, and say in your delivery note that you can re-export to a different spec on request. That single sentence turns a guess into a professional default. For e-learning in particular, the course developer's authoring tool may need a specific format, so the corporate and e-learning narration guide is worth reading before the first job. And if you are choosing or changing editing software, every major DAW can export every spec on this page; the differences are in workflow speed, not in what the finished file can be.

Frequently asked questions

Should I send WAV or MP3 for voice-over?

WAV for any finished job someone will edit, mix or broadcast, because it is uncompressed and loses nothing. MP3 when the destination asks for it (ACX audiobooks, podcast feeds, auditions) or for review copies a client will play on a phone.

Is 44.1 kHz or 48 kHz better for voice-over?

Neither sounds better for voice. Use 48 kHz for anything going to video, broadcast, games or corporate work, and 44.1 kHz for ACX audiobooks, which require it. When in doubt, 48 kHz is the safer default.

What are stems and when should I deliver them?

Stems are separate, unmixed audio elements, for example a raw vocal track without processing alongside the finished, mixed version. Video game studios, corporate clients with in-house editors, and anyone who mentions their own post-production step will often want stems in addition to, not instead of, a final mix. Podcasters with remote guests want the reverse: one stem per person, so each voice can be cleaned and levelled separately.

Should I normalise or limit my voice-over before sending it?

Only if the client wants a finished file or names a target. If they are mixing your voice with music, send it clean with headroom and no heavy limiting. If the destination is a platform with a published target, such as ACX or a podcast feed, deliver to that target.

How should I send large audio files to a client?

Send a single link to a zipped folder in cloud storage or through a file transfer service, or use the client's own portal if they have one. Email attachments are limited (25 MB on personal Gmail), which a few minutes of WAV will exceed.

What is the best file naming convention for voice-over?

Client_Project_Section_Role_Version, with an optional year-first date: for example Acme_Safety_M03_Narrator_v2.wav. Use underscores, leading zeros and plain letters and numbers, and never the word final. If the client supplies a scheme, use theirs exactly.

Sources and Further Reading

Keep reading

Podcasting

How Podcast RSS Feeds Actually Work

11 min read

Niche Work

Temp Voice Over for Animatics: Pay and Risk

13 min read

Technique

The Lombard Effect: Why Noise Makes You Louder

12 min read