Lecture and interview recordings that are too quiet to hear

Open the file in the browser volume booster and press Smart Boost. It measures the whole file and tells you exactly how much gain is left before anything clips, then applies it in one click. That takes about thirty seconds and it's the single biggest improvement you'll get. Then load the five-band equalizer and pick the Voice preset, which pulls down the low-frequency rumble that's masking the speech. Nothing is uploaded anywhere, there's no sign-up, and it runs on the laptop you already have open.

How well it works depends almost entirely on how far you were sitting from the lecturer. Some quiet recordings clean up beautifully. Some don't, and it's worth knowing which one you have.

Why it came out this quiet

Distance is most of the story. Sound from a single source drops about 6 dB every time the distance doubles. Row three is maybe 4 metres from the podium; the back row is 16 metres. That's two doublings, so roughly 12 dB quieter, a quarter of the amplitude, before anything else is accounted for. In a real hall with hard walls the drop is a bit less than the full 6 dB per doubling, because reflected sound fills in behind it. But reflections are smeared and mushy. They add level without adding clarity.

The other thing distance does is change the ratio between the lecturer's voice and everything else. The person coughing two seats over, the projector fan, your own jacket rubbing against the phone: all of that was recorded at close range. The voice you want was recorded from far away. Turning up the file turns up all of it by the same amount. That's the limit nobody selling an "audio enhancer" says out loud.

Measure first, then boost

Before you touch a slider, find out how much room you have. Smart Boost scans every sample in the file and reports four numbers:

• Sample peak in dBFS — the loudest single sample. A quiet phone recording of a lecture often lands somewhere around -28 to -18 dBFS.
• RMS in dBFS — the average energy, which tracks better with how loud the file actually feels.
• Integrated loudness in LUFS — a proper ITU-R BS.1770-4 measurement with K-weighting, 400 ms blocks at 75 % overlap, and both the -70 LUFS absolute gate and the -10 LU relative gate. This is the number streaming platforms use, so if you're re-publishing the lecture it also tells you the gain needed to hit -16 LUFS for Apple Podcasts or -14 LUFS for Spotify. It reports the number and the suggested gain; it doesn't silently re-master the file to a target for you.
• Headroom in dB, plus the exact boost percentage that produces zero clipped samples with 1 dB of margin left below full scale.

Click the suggestion and it applies. One caveat: that peak figure is sample peak, not true peak (dBTP). True peak accounts for what happens between samples when the signal is reconstructed, and it can sit a fraction of a dB higher. For a lecture you'll play back on headphones, sample peak plus 1 dB of margin is plenty. For something you're mastering for release, it isn't.

500 % is the ceiling, and it's worth doing the arithmetic. If your peak reads -20 dBFS you need about 19 dB to get near full scale, and one pass can't reach that. The fix: turn the soft limiter off, apply 500 %, export as 32-bit float WAV, load that file back in, and run Smart Boost again. The 32-bit float part matters. Float keeps the detail and the headroom intact through the round trip, so two passes land where one big one would.

What boosting won't fix

Gain is a multiplier. It has no idea which parts of the waveform are the lecturer and which are the ventilation, so it scales both by the same factor. The signal-to-noise ratio of your file was fixed the moment you pressed record.

Two things are genuinely unrecoverable. If the voice landed near the noise floor of the recording, you'll hear it as a grainy, crunchy quality once you boost it. That grain is noise that was recorded along with the voice, and it comes up with the voice. And if the lecturer was buried under an air handler running in the same frequency range, nothing on this site separates them. These tools have no noise removal and no stem separation. If you can't make out words, your better moves are the university's own lecture capture system, a classmate's recording, or the slides plus a friend's notes.

EQ does more for speech than gain does

Intelligibility and loudness are different problems. You can make a file measurably louder and still not understand it. What makes speech understandable is the consonants, the t, k, s and f sounds that separate "fifteen" from "fifty", and those live roughly between 2 kHz and 5 kHz. Vowels carry the volume. Consonants carry the meaning.

The Voice preset in the five-band equalizer is built for exactly this. Its bands are -4 dB at 60 Hz, -2 dB at 230 Hz, +3 dB at 910 Hz, +5 dB at 3 kHz and +2 dB at 14 kHz, each one a peaking filter with a Q of 1.0. The 60 Hz cut takes out ventilation rumble and the thud of desks and footsteps. The 230 Hz cut removes the boxy, muddy build-up you get in a room with hard surfaces. The +5 dB at 3 kHz is the band doing the real work: that's consonant territory, and lifting it can move a recording from "I catch every third word" to fully followable with no change in measured level at all.

Use the level-matched A/B button while you adjust. It flips between processed and untouched and trims the processed side down to the original's peak, so you're judging tone instead of being fooled by loudness. Louder always sounds better for the first two seconds. Matched-level comparison is the only way around that, and no other free browser tool I know of offers it.

Skip the bass booster for lectures. It's a low shelf at 120 Hz, and for spoken word that region holds almost nothing you want and quite a lot of rumble you don't.

The limiter, and the lecturer who paces

Professors walk. They turn to the whiteboard mid-sentence, lean into the front row for a question, walk back. Your recording swings 10 dB or more inside one paragraph, and if you set the gain so the loud moments stay clean, the quiet moments stay unusable.

The soft limiter is on by default and it helps here. It's a tanh-knee waveshaper: below 0.7 of full scale it does essentially nothing, and above that it curves asymptotically toward full scale instead of chopping the tops off. So you can push the gain past what the peaks strictly allow, and the loudest moments get rounded rather than shattered into distortion. It's a limiter, not a compressor with attack and release, so it won't ride the whole lecture up and down. What it does is let you choose a boost that suits the quiet passages without the loud ones turning to buzz.

Practical recipe: run Smart Boost, note the percentage it gives you, then push the slider another 20 to 40 points past it with the limiter on. Watch the live clip warning, which tells you what share of the file would run past full scale at the current setting. On speech, a fraction of a percent going into the limiter's knee is not something you'll hear. If it's climbing through several percent, back the slider off.

Playback tricks that help comprehension

Once you have a usable file, slow it down. Most students race through recordings at 1.5x or 2x, but with a marginal one the opposite works better. At 0.85x or 0.9x your brain gets more time per syllable to reconstruct the consonants it half-heard, and any modern player does this without shifting the pitch. It costs about ten extra minutes across an hour and it can be the difference between guessing and knowing.

Headphones beat speakers every time, and closed-back ones also shut out the room you're in. A laptop speaker is too small to reproduce much below about 500 Hz, where the fundamentals of the voice sit, so it hands you a thin version of an already-thin recording. Play the boosted file through once before you start taking notes, too. Then you know which five minutes are the hard part and can save your attention for them.

Record the next one properly

All of that is damage control. Ten seconds of setup next week beats an hour of rescue work.

1. Sit closer. Moving from 12 metres to 6 metres buys you around 6 dB for free, and it costs nothing but arriving five minutes earlier.
2. Get the phone off the desk and off your lap. A desk conducts every pen tap and knee bump into the mic. Prop it upright against a bag with the mic end pointing at the lecturer; on most phones that's the bottom edge, next to the charging port.
3. Take the case off, or at least check that it isn't covering the mic port. This one silently ruins more recordings than anything else.
4. Ask about the room mic. Plenty of halls already record to the LMS, and if the lecturer wears a lapel mic there's a clean feed you're duplicating badly from 20 metres away.
5. Do a 20-second test in the first minute, play it back, and move if it sounds thin. Fixing it at minute one is free.

If most of your recording happens on your phone and you'd rather not shuttle files to a laptop afterwards, the same five EQ bands and the same preset values are in the Increase Volume app for iPhone and the Android version, both of which keep a library of your files with settings saved per track and work offline. The app tops out at 200 % rather than 500 %, and it has no limiter and no level-matched A/B, so for a difficult file the browser tools still have the edge.