search⌘KtranslateENexpand_moreupload_fileConvert

Record your voice, check the level, save the take.

Press record and speak. The meter shows whether you're too quiet or distorting while you talk. Keep several takes, listen back, and download the one you want as WAV or a small compressed file. Audio never leaves your device.

ready0:00.0

−60 dB−25−70 dB

pausemic

mic_off

microphone processing · set before recordingNoise suppressionRemoves steady background hum, fans and traffic.Echo cancellationStops speaker sound feeding back into the mic.Auto gainTurns the volume up and down to keep speech even.

For speechall onFor musicall off

Saved in the browser’s own format (usually Opus, ~24 KB per 10 s). WAV is lossless, about 1 MB per 10 s.

takesnone yet

graphic_eqEach recording appears here with a player and download buttons. Takes stay until you close the page.

downloadWAV · · delete

~/reading-the-meter

Aim for the green. Red can't be fixed later.

Digital audio has a hard ceiling at 0 dB. Anything louder is cut flat, and that crackle stays in the file. Too quiet is safer, but turning it up later also turns up the hiss. The three zones on the meter mean this:

Too quietbelow −25 dBSpeech will be hard to hear and boosting it later brings up room noise and hiss.Move closer, 10-20 cm from the mic.

Good−25 to −7 dBLoud enough to be clean, with space left for a laugh or a raised voice.Keep this distance and speaking level.

Too loudabove −7 dBPeaks near 0 dB are clipped: the tops of the waves are cut off, and it crackles.Back off or turn the input level down.

the zone you’re in lights up while recording

~/voice-or-music

The same three switches, set for four jobs.

Your browser's microphone processing was designed for video calls. It makes voices clearer, but it can squash music and room sound. Click a job to apply its settings above.

record_voice_overVoice note

noiseonechoonautoon

Clearest speech with no setup. Fine for messages and reminders.

podcastsPodcast or voiceover

noiseonechooffautooff

Keep noise suppression but set your own level, so quiet pauses don’t pump up the hiss.

pianoMusic or singing

noiseoffechooffautooff

All off. Suppression treats sustained notes as noise and chops them.

forestRoom or nature sound

noiseoffechooffautooff

All off, to capture the space as it is. Expect a lower level.

~/what-you-asked-for-vs-what-you-got

The switches are requests. The browser decides.

When this page opens the microphone it asks for noise suppression, echo cancellation and auto gain on or off. The browser may ignore a request it can’t honour, run at a different sample rate, or record in a format of its choosing. Everything below is read from your browser now, and from each take after you record it.

Settings this browser understands

noiseSuppressionNoise suppression switchsupported

echoCancellationEcho cancellation switchsupported

autoGainControlAuto gain switchsupported

voiceIsolationNewer, stronger voice filtersupported

sampleRateAsk for a sample ratesupported

sampleSizeAsk for a bit depthsupported

channelCountMono or stereosupported

latencyAsk for low delaysupported

deviceIdChoose a specific micsupported

navigator.mediaDevices.getSupportedConstraints()

What your last take actually used

Noise suppression--

Echo cancellation--

Auto gain--

Sample rate--

Channels--

Recorded as--

Record a take above and this column fills in from MediaStreamTrack.getSettings().

Formats this browser can record

audio/webm;codecs=opusyes

audio/webmyes

audio/ogg;codecs=opusno

audio/oggno

audio/mp4;codecs=opusyes

audio/mp4;codecs=mp4a.40.2yes

audio/mp4yes

audio/mpegno

audio/wavno

audio/webm;codecs=pcmyes

Audio engine runs at 48,000 Hz · base latency 10.0 ms. MediaRecorder.isTypeSupported()

~/how-small-is-too-small

Five bitrates, one voice, measured.

A 4-second synthetic voice, a buzzing 120 Hz source through moving vowel formants, plus hissy “s” sounds up to 10 kHz: is played into five recorders at once, each asked for a different bitrate. The files are then decoded and their spectra measured. Low bitrates don’t just sound muddier: the encoder throws away everything above a cut-off frequency.

Asked forGotFileAudio up to

OriginalPCM375 KB…uncompressed WAV

6 kbps………

12 kbps………

24 kbps………

48 kbps………

96 kbps………

Rows fill in after the test. The synthetic voice is generated in your browser; no microphone is needed.

average spectrum · dB0-20 kHz

graphic_eqPress “Run codec test”, it plays a 4-second synthetic voice silently into five recorders and takes about 5 seconds.

~/reference

Inside the WAV this page writes.

A WAV file is a 44-byte header followed by raw samples. These are the exact bytes for the 4-second synthetic voice used above, decoded field by field, with the sizes other common formats would take for the same length.

52 49 46 46 24 DC 05 00 57 41 56 45 66 6D 74 20 10 00 00 00 01 00 01 00 80 BB 00 00 00 77 01 00 02 00 10 00 64 61 74 61 00 DC 05 00

OffsetBytesFieldValue

04Chunk ID“RIFF”

44Size of the rest of the file384,036 bytes

84Format“WAVE”

124Sub-chunk ID“fmt ”

164fmt chunk size16

202Audio format1 = PCM, uncompressed

222Channels1 (mono)

244Sample rate48,000 Hz

284Byte rate96,000 bytes/s

322Block align2 bytes per frame

342Bits per sample16

364Sub-chunk ID“data”

404Sample data size384,000 bytes

The same 4.0 s in other formats

Telephone8 kHz · 8-bit µ-law · mono31 KB

Speech recognition16 kHz · 16-bit · mono125 KB

This page’s WAV48 kHz · 16-bit · mono375 KB

CD audio44.1 kHz · 16-bit · stereo689 KB

Studio48 kHz · 24-bit · stereo1.2 MB

High resolution96 kHz · 24-bit · stereo2.3 MB

MP3 at 128 kbpstypical, lossy63 KB

Opus at 24 kbpstarget, lossy12 KB

Timeline

1877Edison records sound on tinfoil with the phonograph.

1928Nyquist shows a signal must be sampled at twice its highest frequency.

1937Alec Reeves invents pulse-code modulation (PCM).

1972ITU G.711 standardises 8 kHz µ-law and A-law telephone audio.

1982CD audio fixes 44.1 kHz, 16-bit stereo PCM.

1991IBM and Microsoft publish RIFF and the WAVE format.

1993MP3 (MPEG-1 Audio Layer III) is standardised.

2012Opus becomes IETF RFC 6716.

2013MediaRecorder ships in Firefox; Chrome follows in 2016.

2017WebRTC 1.0 constraints include noiseSuppression, echoCancellation and autoGainControl.

2021Safari ships MediaRecorder, recording to MP4/AAC.

2024voiceIsolation is added to the Media Capture spec as a separate constraint.

Sources

W3C: Media Capture and StreamsgetUserMedia, constraints and getSettings.W3C: MediaStream RecordingMediaRecorder, mimeType and audioBitsPerSecond.W3C: Web Audio APIAudioContext, OfflineAudioContext and decodeAudioData.IETF RFC 6716: OpusBandwidths from narrowband to fullband and how bitrate selects them.IETF RFC 7845: Ogg OpusHow Opus is stored in Ogg files.Microsoft & IBM (1991): RIFF / WAVEMultimedia Programming Interface and Data Specifications 1.0.ITU-T G.711Pulse code modulation of voice frequencies.Nyquist (1928), Trans. AIEE 47(2)Certain topics in telegraph transmission theory.

~/after-recording

Trim it, join it, shrink it.

What people usually do with a recording next.

content_cutAudio clipperCut the silence at the start and end of a take.1:42 takearrow_forward1:31 clip queue_musicAudio joinerCombine several takes into one file.3 takesarrow_forwardjoined.wav compressWAV to MP3Shrink the WAV for email or messaging.take-3.wav · 10 MBarrow_forward.mp3 · 1 MB