Speech denoising: hear the input and two actual results
The same 30-second speech input was processed with MeluMimi’s standard and gentle settings. Listen for background noise, speech detail and artifacts. These are actual outputs, not manually cleaned replacements.
Compare the same speech excerpt
Speech reference
The licensed source excerpt before our added noise; not a noise-free studio master.
30 seconds · 48 kHz · mono · 16-bit WAV
Download this WAVInput with added background noise
A controlled input: reproducible synthetic hiss and hum were added to the reference.
30 seconds · 48 kHz · mono · 16-bit WAV
Download this WAVStandard denoising
The actual standard-setting output, with no extra EQ, gain adjustment or manual cleanup.
30 seconds · 48 kHz · mono · 16-bit WAV
Download this WAVGentle denoising
The actual gentle-setting output. Compare retained ambience and speech with the standard result.
30 seconds · 48 kHz · mono · 16-bit WAV
Download this WAVSource, changes and processing
Speech: LibriSpeech © 2014 Vassil Panayotov, CC BY 4.0; prepared with Daniel Povey. Reader: Nolan Fout, “Ten Years Later”, chapter “11 - Night”; records 6930-75918-0004 through 0007. We concatenated those records in order, took the first 30 seconds and resampled the 16 kHz source to 48 kHz mono. We added seeded synthetic white noise and 60/120 Hz hum at a nominal 5 dB signal-to-added-noise ratio, then applied one shared input gain to avoid clipping. Both settings used the current MeluMimi processing wrapper and the same private runtime artifact as production, on a local Mac CPU. This was not a website upload task or an online speed test. The output WAVs are unchanged; browser MP3s are 320 kbps playback encodes.
Read the source and licence · Read CC BY 4.0
Source excerpt and these adapted example audio files are provided under CC BY 4.0. The reader and dataset authors do not endorse MeluMimi. This page contains no customer recordings.
How to denoise your own speech recording
- Open the audio denoiser and sign in. Choose audio or video you own or are authorized to process.
- Choose standard or gentle. Preview the first 60 seconds without consuming paid minutes.
- Listen to speech and quiet sections. If full processing is suitable, review and confirm its paid-minute cost before starting.
- After full processing, download the WAV and keep your original recording. Full processing charges the whole source duration, including the previewed portion.
What this example does not prove
This is one English read-speech excerpt with added synthetic noise, not a real interview, a benchmark or a universal quality score. The reference can contain its own recording noise. Noise reduction can change voice details; music, wind, clipping and overlapping speakers may behave differently. Resampling does not restore details missing from the source. Public example downloads are separate from the tool’s paid full-processing workflow.