Audio Volume Normalizer

How do you normalize the volume of an audio recording?

Choose audio files, select a target under Normalization settings, and press Preview levels. The tool measures the original recording and calculates a fixed gain for each file. Review Before and After, then press Build WAV files to prepare downloads. Every result has its own Output filename and Download WAV action. You can also use Copy measurements to keep a text summary of the level change.

Peak normalization is the starting mode, with a target of −1 dBFS and no extra gain. It places the largest absolute sample near that level while applying the same multiplier to every channel. It does not change the relative level of quiet and loud passages within a recording. If a speaker moves away from the microphone halfway through a take, that section stays quieter relative to the rest. Normalize after choosing the useful passage with Audio Cutter when unwanted sections would otherwise influence the measurement.

What happens in a worked peak normalization example?

Consider a recording whose largest absolute sample is 0.5. Its sample peak is approximately −6.02 dBFS because the tool calculates decibels as 20 × log10(amplitude). With Target (dBFS) set to −1 and Extra gain (dB) set to zero, the applied gain is approximately +5.02 dB. Multiplying every sample by about 1.7825 brings that largest sample to approximately 0.89125, the linear amplitude corresponding to −1 dBFS.

If the original RMS is 0.25, approximately −12.04 dBFS, the same gain raises it to about 0.44563, approximately −7.02 dBFS. Peak mode does not independently choose an RMS result: both measurements change together under the fixed multiplier. The screen displays decibels to two decimal places and linear amplitudes to six decimal places, so rounded values may not reproduce the internal result exactly.

Now add +6 dB of Extra gain without changing the target. The predicted largest sample becomes about 1.77828, above the representable full scale of the output. Preview levels reports samples that would clip, and Build WAV files stays unavailable while the processed result contains samples outside ±1. Reduce Extra gain, lower the target, or enable Hard limiter and preview again. Turning on the limiter at −1 dBFS caps the affected samples near 0.89125; it does not preserve their original shape.

Should you choose peak normalization or an RMS target?

Choose peak mode when your immediate goal is a defined sample ceiling using one unchanged gain throughout the file. Choose RMS when you want to compare an unweighted average signal level across separate recordings. Selecting a different mode resets the target to its initial value, so check Target (dBFS) after switching. Both modes accept values from −60 to 0 dBFS. Extra gain is added afterward and ranges from −24 to +24 dB.

RMS includes all samples, including silent gaps. Two otherwise identical recordings with different amounts of silence therefore receive different RMS gains. The measurement has no frequency weighting or silence gate and is not a perceptual loudness measurement. Matching RMS targets does not guarantee that speech, music, and background noise sound equally loud. There is no LUFS meter or loudness standard compliance claim here.

How should you interpret clipping and limiter measurements?

Would clip counts individual channel samples whose predicted magnitude exceeds 1 before limiting. A stereo frame contains two channel samples, so this count is not a duration or a count of moments in time. Samples exactly at +1 or −1 are not counted as clipping. Predicted peak before limiter shows the largest projected linear magnitude before any ceiling is applied. Limited counts samples capped by Hard limiter, which can include samples below full scale when its ceiling is negative.

The limiter ceiling accepts −24 to 0 dBFS and starts at −1 dBFS. It is a hard sample limiter: each sample above the ceiling is replaced with the ceiling magnitude while retaining its sign. It has no lookahead, oversampling, attack, or release settings. Limiting can change channel ratios at affected samples and introduce distortion. After reports the actual processed measurements, even when those values miss the requested RMS target. Listen to the downloaded result before using it in another edit.

All After measurements describe processed Float32 PCM before conversion into integer WAV samples. The 16-bit or 24-bit encoder then quantizes those samples, introducing small numerical differences. These values are sample peaks rather than estimates of reconstructed true peaks. For an entirely silent input, peak and RMS appear as Silence (−∞ dBFS); the tool applies zero gain and exports silence rather than trying to amplify a nonexistent signal.

Which file sizes, formats, and batch limits apply?

Choose audio files accepts common audio extensions, including MP3, WAV, M4A, AAC, OGG, Opus, FLAC, and WebM. Actual decoding depends on the browser codec implementation. A filename extension alone does not establish support, and changing an extension does not convert a recording. An unsupported or corrupt input produces a text error. If necessary, create a genuine WAV copy using a local audio application and select that copy.

Each selected file must be nonempty and at most 104,857,600 bytes, or 100 MiB. A batch holds at most 20 files and 8,388,608 decoded channel samples, or 33,554,432 bytes of Float32 PCM. For example, stereo audio decoded at 44,100 Hz uses 88,200 channel samples per second, so the combined sample budget holds about 95.1 seconds at that configuration. Mono at the same rate uses half as many samples per second. This budget is about retained PCM, not the compressed size shown in a file manager.

Files are decoded sequentially and checked against the remaining batch budget. However, the browser must decode an entire file before the tool can inspect its sample count, and workers can create temporary copies. The 32 MiB budget is not a guarantee about peak memory usage. The accepted decoded channel count is 1–8, and decoded integer sample rates range from 8,000 to 192,000 Hz. An error rejects the batch rather than publishing a partially successful set. Choose replacement files to retry with a smaller selection.

What does WAV output preserve and what stays on your device?

Output offers WAV 16-bit and WAV 24-bit. Processing preserves the duration, channel count, and sample rate of the decoded audio; it does not offer resampling or original codec passthrough. The shared browser decoder may resample input while decoding, so inspect the decoded Hz label instead of assuming it matches source metadata. Stereo channels share one gain per file, while different files receive independent gains. There is no shared album gain or automatic channel balancing.

Uncompressed WAV size depends on frames, channels, and bytes per sample. A ten second stereo recording decoded at 44,100 Hz needs 1,764,000 PCM bytes at 16-bit or 2,646,000 at 24-bit, plus the WAV header. Selecting 24-bit does not restore detail removed by a compressed source. Each WAV downloads separately; ZIP packaging and compressed audio export are outside this version.

Sources and generated WAV files remain in device memory until you save or explicitly share them. There is no saved browser draft or server copy. Cancel processing stops the worker, while browser decoding may finish internally before its cancelled result is discarded. Adjust settings returns to the original decoded files and clears the previous export. Normalize more files releases the current batch from the interface. Start with Voice Recorder for a new microphone take, or use File Hash Verifier when you need to compare a downloaded recording against an expected checksum.