Which sample rate and bit depth should you use for voice?

A one-hour interview recorded in mono at 96 kHz and 32-bit float takes about 1.4 GB on disk. The same hour at 48 kHz and 24-bit takes about 520 MB. Both files then get exported to a 128 kbps MP3 for the podcast feed, and both MP3s weigh the same 58 MB. Everything the bigger file captured above 24 kHz disappears in that export: an MP3 at 44.1 or 48 kHz has no room for it.

That is the whole argument in one example. Sample rate and bit depth matter, but for a speaking voice the sensible values were settled long ago, and pushing them higher mostly buys you bigger files. What actually changes how you sound is somewhere else: mic distance, the room, input level, noise.

Here is what to pick, why, and where each setting hides on a Mac.

The short answer: 48 kHz, 24-bit

The best sample rate for recording a speaking voice in 2026 is 48 kHz, and the bit depth to pair it with is 24. Set the input to 48 kHz, 24-bit and move on. The exceptions are narrow.

If you only came here to find the menu where this gets changed, the guide to Audio MIDI Setup on a Mac walks through that window field by field, and that page is the better read for you. The rest of this one explains the numbers, so you can tell when a tutorial that says to record at 192 kHz is just eating your disk.

What the audio sample rate actually decides

Digital audio is a series of snapshots of the waveform. The sample rate is how many snapshots per second: 48 kHz means 48,000 of them. A sampled signal can only hold frequencies up to half its sample rate. The Audacity manual puts it plainly for the CD rate: 44,100 Hz reproduces frequencies up to 22,050 Hz, which covers what people can hear.

So the sample rate sets the ceiling on how high a sound can go and still be recorded. It says nothing about how clean, warm, or clear the sound below that ceiling is.

Now look at where a voice lives. The fundamental of a speaking voice sits in the low hundreds of hertz. The consonants, the hiss of an "s", the air around the words reach several kilohertz higher. A traditional phone call is sampled at 8 kHz, which cuts everything above 4 kHz, and people have understood each other on phones for decades. A Bluetooth headset mic on a call usually runs over the hands-free profile, where the classic codecs work at 16 kHz (wideband) or 8 kHz (narrowband). That is why a headset mic in a call can sound like a phone: it is, technically, close to a phone.

Between 16 kHz and 48 kHz is where "sounds like a phone" turns into "sounds like a person in the room". Above 48 kHz there is nothing a listener gains from a speaking voice. 96 kHz captures content up to 48 kHz, which is more than twice the top of human hearing.

44.1 vs 48 kHz

Both rates cover the full audible range, so the choice is about where the file is going, not about quality.

• 44.1 kHz is the CD standard and the Audacity default. Music distributors and older audio tools assume it.
• 48 kHz is the standard for video. YouTube recommends it for uploads, OBS switched its default to it in version 26.0, Discord's voice protocol expects Opus audio at 48 kHz, and Zoom's high-fidelity music mode raises its codec to 48 kHz.

The real risk is not picking the "wrong" one. The risk is mixing them. When your mic delivers 44.1 kHz and the app records at 48 kHz, something in between has to resample the audio on the fly. Done well, you will never notice. Done badly, or across two devices with separate clocks, it produces clicks, crackles, or a slow drift between sound and picture over a long recording. Apple's guide to aggregate devices in Audio MIDI Setup makes the same point: ideally every device runs at the sample rate of the one chosen as the sync source, and devices that cannot are kept in step with drift correction.

So pick one rate for the whole chain and keep it. For voice on a Mac in 2026 that rate is 48 kHz, because so much of what you will feed it into (calls, video, streams) already runs there.

What bit depth actually decides

Bit depth is how many steps each snapshot can use to describe the level of the wave. More bits mean finer steps, and finer steps mean the rounding error (quantization noise) sits lower. Each extra bit buys about 6 dB of range.

Bit depth does not add "detail" to a voice the way people expect. What it gives you is room. With 24 bits you can record your voice peaking around -12 dBFS, leave a wide safety margin for the laugh or the cough that comes out louder, and the noise of the format itself still sits far below anything your room or preamp produces. At 16 bits the same safety margin eats into the quiet end. Not dramatically, but there is no reason to accept it when 24-bit costs nothing except disk space.

Does 32-bit float save you from clipping?

Partly. 32-bit float earns its place in editing: it is Audacity's default, and inside a float project a gain boost or an effect has so much headroom that it does not clip the stored samples. But the clipping you hear on a voice usually happens earlier, in the mic or interface converter, which outputs integer samples. If your shout was too loud for that converter, the top of the wave was already cut off before any software saw it, and storing the result as float does not grow it back.

Some field recorders get around this with two converters at different gains and 32-bit float recording from the input onwards; Zoom Corporation (no relation to the call app) builds its F-series recorders that way. Those can recover overloads that integer recorders cannot. A typical USB mic on a desk works differently. For a regular setup, the protection against clipping is still proper input gain, which is a separate topic covered in the article on gain staging.

Is 96 kHz or 192 kHz better for voice?

For speech, no. The extra sample rate stores frequencies no listener hears, and every downstream step (call codec, YouTube, podcast MP3) resamples the audio down again. What you do get is bigger files and more processing work:

One hour of mono WAV (decimal megabytes)

44.1 kHz / 16-bit integer     317.5 MB
48 kHz   / 24-bit integer     518.4 MB
48 kHz   / 32-bit float       691.2 MB
96 kHz   / 24-bit integer   1,036.8 MB
96 kHz   / 32-bit float     1,382.4 MB
192 kHz  / 32-bit float     2,764.8 MB

Stereo doubles every line.
128 kbps MP3 of the same hour:  57.6 MB

There are niches where high rates earn their keep: sound designers who pitch recordings down by an octave or two, people recording ultrasonic sources, some mastering workflows. A voice for a call, a lesson, a stream or a podcast is not one of them.

Which sample rate for a podcast

Apple publishes its audio requirements for Apple Podcasts, and they are relaxed. For WAV and FLAC uploads, the minimum is 44.1 kHz at 16 or 24-bit, and the recommendation is 24-bit at any of 44.1, 48, 88.2, 96, 176.4 or 192 kHz. For MP3 in the feed, Apple recommends 44.1 or 48 kHz, 96 to 128 kbps for mono and 128 to 256 kbps for stereo, with overall loudness around -16 LKFS and true peaks below -1 dBFS.

So the best sample rate for a podcast is the one you can keep consistent from mic to export:

1. Record at 48 kHz, 24-bit. If you also publish video clips of episodes, you are already on the video rate.
2. Edit at the same rate. Set the project rate in your editor to match before you hit record, not after.
3. Export the delivery file to whatever your host asks for. If the host wants 44.1 kHz MP3, the conversion happens once, at the end, where it costs nothing.

Recording at 44.1 kHz is equally fine if your whole chain is audio-only. The one habit to avoid is switching between the two inside a project.

If you are just starting to record and the question is really "which app, which settings, which mic", the guide to recording audio on a Mac covers the full chain, not just the format.

Where to change sample rate and bit depth on a Mac

macOS has one system-wide place for the hardware format, and then each app has its own idea of what it records.

Audio MIDI Setup: the hardware format

Open Audio MIDI Setup from Applications > Utilities. In the Audio Devices window, select your microphone in the sidebar, then use the Format pop-up menu on the right to set the sample rate and bit depth. Apple's guide adds one caveat: the values should match what your device supports. A USB mic will only list the formats its own converter offers, so do not be surprised if the menu holds just one or two lines.

This is the setting to check first when a recording crackles or drifts. A full walk through the window, including aggregate devices, is in the Audio MIDI Setup guide.

QuickTime Player and Voice Memos

QuickTime Player's audio recording has two choices in the Options pop-up menu, according to Apple: High, which records 44,100 Hz AAC, and Maximum, which records 44,100 Hz uncompressed 24-bit Linear PCM. For a voice take you plan to edit, choose Maximum.

Voice Memos keeps it simpler: Voice Memos > Settings has an Audio Quality choice between Compressed and Lossless. Lossless for anything you will reuse.

Audacity

In Audacity on a Mac, the settings are under Audacity > Preferences > Audio Settings. The default sample rate is 44,100 Hz and the default sample format is 32-bit float. If your mic and the rest of your work run at 48 kHz, change the project rate to 48,000 Hz before recording, so Audacity does not resample every take.

OBS

In OBS, go to Settings > Audio > General and set the sample rate. 48 kHz has been the default since OBS 26.0. Restart OBS after changing it, so the new rate takes effect for the next recording. For the rest of that panel, from channel layout to the mic and desktop audio sources, see OBS audio settings on a Mac.

Zoom

Zoom chooses its own codec settings for normal calls. The one switch that touches this topic is in Settings > Audio: under microphone modes, Original sound for musicians with High-fidelity music mode raises the codec to 48 kHz, 96 kbps mono or 192 kbps stereo, according to Zoom. It also turns off echo cancellation and post-processing, and Zoom says it is meant for a professional interface, microphone and headphones. For a normal spoken meeting, leave it off; the mode is built for instruments.

What a call does to your perfect settings

Here is the part tutorials skip. When you are on a call, your 48 kHz, 24-bit signal goes into the app, gets cleaned up by the app's own processing, and then gets compressed by a codec to fit the network. Browser calls run on WebRTC, whose spec requires Opus and G.711 support; Opus itself works from 8 kHz up to 48 kHz depending on the bandwidth it has. Discord's voice protocol uses Opus at 48 kHz. Whatever you set in Audio MIDI Setup above 48 kHz is resampled down before it leaves your Mac, and the bit depth stops meaning much once a lossy codec takes over.

And if the mic is a typical Bluetooth headset in call mode, the ceiling is the hands-free profile's 16 kHz or 8 kHz, whatever the format menu says. No format menu fixes that. A wired or USB mic does. The article on Bluetooth headsets for calls goes through which ones handle this better.

So on a call, people hear the difference between 44.1 and 48 kHz far less than they hear:

• a mic an arm's length away instead of a hand's width;
• a room that rings like a bathroom;
• a level that is too quiet, so the app's gain pushes the noise up with it;
• a laptop fan, a dishwasher, a hair dryer in the next room;
• a mic whose tone is thin or boomy for your particular voice.

Those are signal problems, and no sample rate touches them. The free fixes come first: move closer, set the gain properly, put something soft behind you, turn on the noise suppression your call app already has. The guide on making your mic sound better without buying a new one goes through them in order.

Where TunedMic fits

If you have done the free part and your voice still comes across thin, noisy, or echoey on calls and recordings, that is the gap TunedMic was built for. It adds a virtual microphone to the Mac. Your real mic stays where it is; TunedMic takes its signal, processes it on the Mac in real time, and any app you point at the TunedMic mic (Zoom, Meet, Teams, Discord, OBS, Telegram, a browser) gets the processed voice.

The setup is a one-minute Sound Check: you read a short passage aloud and TunedMic takes more than 30 measurements of your voice, your mic and your room. Nine profiles, each built from that measurement, cover calls, streams and singing, and switching profile changes the sound in every app at once. Manual controls for noise, room echo, clarity, loudness and a 12-band equalizer are there if you want them, and you never have to touch them. It adds less than a tenth of a second of delay and runs entirely on the Mac; the voice is not uploaded.

The limits, just as plainly. TunedMic does not change your sample rate or bit depth, does not make a Bluetooth headset escape the hands-free profile's limits (a wired or USB mic gives it more to work with), and does not remove every noise. It needs a Mac with Apple silicon (M1 or newer) and macOS 14 or later; there is no Windows or Intel version.

To see the setup before installing anything, watch the 38-second video on the home page: the one-minute Sound Check, then the tuned mic at work in Zoom, Google Meet and Discord. The app costs $19 once, with a free 7-day trial that starts at first launch, opens every feature and needs no card or account. There are no refunds, so use the trial on your real mic and your real calls. Details are on the pricing page.

A settings chain you can copy

Mic / interface (Audio MIDI Setup > Format)   48,000 Hz, 24-bit
Recording app project rate                    48,000 Hz
Recording format                              WAV or AIFF, 24-bit (or 32-bit float in Audacity)
Peak level while talking                      around -12 dBFS, never touching 0
OBS (Settings > Audio > General)              48 kHz, restart after changing
Podcast export                                MP3 44.1 or 48 kHz, 96-128 kbps mono,
                                              about -16 LKFS, true peak below -1 dBFS
Video export                                  AAC, 48 kHz
Calls                                         leave the app on its defaults

Check this chain once, and the format stops being a variable. If something still sounds wrong after that, the cause is in the signal, not in the numbers.

Your next step

Open Audio MIDI Setup now, select the mic you actually use, and look at the Format menu. If it says 44.1 kHz while your editor or OBS runs at 48 kHz, set both to 48 kHz and record thirty seconds in the app you normally record with. Listen back on headphones. If the take is clean but your voice is thin, roomy or buried in fan noise, run the 7-day trial against your own mic on your next real call.