Why speaker levels vary
Almost every conversation recording has some level mismatch between the people in it. The usual causes are ordinary:
- Different microphones and different input gain for each person.
- Distance to the microphone. A host who leans in is several decibels louder than a guest who sits back, and people move while they talk.
- Remote calls. Each participant’s software and hardware sets its own level, and a guest on laptop speakers or earbuds is often quiet.
- Some voices are simply louder than others.
The result is a mixed file where the listener keeps reaching for the volume control: up for the quiet guest, down when the host comes back.
Why normalizing alone does not fix it
Normalization, whether to a peak level or to a loudness target, applies one gain to the entire file. Everyone gets louder or quieter by the same amount, so the gap between the loud speaker and the quiet speaker stays exactly as it was. Normalizing is still worth doing, but it answers a different question: how loud is the whole episode.
A compressor can narrow the gap, but it also squeezes the natural rise and fall inside each person’s speech, and it needs threshold, ratio, and timing settings that suit the material. Adjusting clip gain by hand in an editor works, but a long interview can have hundreds of handoffs.
How Pod Level approaches it
Pod Level works on the conversation as a sequence of speech turns rather than as one block of audio:
- It detects speech and splits it into turns using a small voice activity model that runs in your browser.
- It measures the loudness of each turn.
- It finds the recording’s typical speech level (a duration-weighted median across turns) and moves each turn toward it, by up to 12 dB in either direction. Very short or uncertain turns follow the gain of the speech around them, and gains change with a short ramp at turn boundaries.
- It then normalizes the whole programme to −16 LUFS and limits true peaks to −1 dBTP so the file is ready to publish.
Within a turn the gain is steady. Pod Level does not compress the voice itself, so emphasis, pauses, and laughter keep their shape; only the level of each turn changes. After processing it shows the measured input and output loudness, the range of gain it applied, and how many turns it adjusted, and lets you preview a few handoffs before and after.
How to balance the speakers in a recording
- Open Pod Level in current iPhone Safari, Chrome on iPhone, or desktop Chrome or Edge.
- Choose the episode file. WAV, MP3, M4A/AAC, OGG or Opus, WebM, and FLAC inputs are accepted. The file is opened read-only and is never uploaded.
- Review the duration and the proposed output format. You can keep the proposed format or choose a smaller compressed one.
- On desktop, pick where to save the new file and start. On iPhone, the result is offered for download or sharing when it finishes.
- Wait for analysis and rendering. Progress, speed, and an estimate are shown, and you can cancel at any point.
The original recording is not changed. If the balance is not what you want, nothing is lost.
What Pod Level does not do
It levels one mixed conversation file. It does not remove noise, apply EQ, duck music, or work on separate tracks per speaker. If your recording is inconsistent over time rather than between people, see fixing uneven podcast volume. If you used to rely on The Levelator for this job, see the Levelator alternative page.