How to Clean Up Lecture and Talk Recordings (Remove Music & Noise)
Some of the most valuable audio in the world is a recording someone made on a phone: a lecture, a khutbah, a class, a grandparent telling a story. And so much of it is marred by background music playing in the room, traffic, or a soundtrack someone added years ago. Here's how to rescue those recordings.
Step 1: Extract the voice
Upload the recording to a free AI voice extractor. The model isolates human voices and discards music and most background sound. This single step fixes the most common problem — music under speech — better than any traditional filter, because the AI understands what a voice is rather than just cutting frequencies.
What it keeps: every human voice in the recording, including the audience's. What it removes: music, instruments, most mechanical and ambient noise.
Step 2: Judge the result honestly
Play the extracted voice next to the original. In most cases you're done. Two situations need extra care:
- Very quiet speaker, very loud music: some voice detail is lost with the music; the result is intelligible but thin. This is the physics of the situation, not a tool failure.
- Question-and-answer recordings: distant audience voices are preserved but stay distant. Voice extraction doesn't re-balance volumes.
Step 3 (optional): polish the speech
For recordings that matter, ten minutes in the free editor Audacity goes a long way after extraction:
- Normalize loudness so the whole recording sits at a comfortable, even level (Effect → Loudness Normalization, −16 LUFS is a good target for speech).
- Gentle compression (Effect → Compressor, ratio ~2:1) evens out the difference between the speaker leaning into and away from the mic.
- A high-pass filter at ~80 Hz removes room rumble and handling noise the extraction may have kept (it can ride along with voices).
Export as MP3 at 128 kbps or higher — plenty for speech.
Archiving family and community recordings
If you're digitizing a collection — a community's lecture archive, a family's voice memos — process the originals and keep both versions. Storage is cheap; a future tool will always be able to do more with the untouched source than with an already-processed copy.
The whole workflow in one line
Original file → voice extraction → quick listen → (optional) normalize and compress in Audacity → share a recording people can finally hear clearly.