THE BRIEF
We need music to sit under our narration for a 45-minute episode. Tense but not distracting. Our host has a low voice, and listeners complain they can't hear her over the music.
Scenario: Host of an independent true-crime podcast
Package: Background bed for an episode · $35 · 24 hrs
The intro track is "low_cello_pulse": 0:00 to 0:20 alone, and the host comes in at 0:20 over a single held note. Nothing moves in her range while she speaks.
THE PROBLEM WITH THE OLD BED
Your host's voice sits around 150 to 400 Hz. The old music had a cello playing right there, so the two fought. I built the new bed with space in that range.
THE BED
- Tempo: 70 bpm, slow pulse.
- Sound: low synth pad below 120 Hz, a soft high piano above 2 kHz. Almost nothing between, where her voice lives.
- Loudness: mixed to sit 18 dB under the voice. After your final mix, the full episode should land at -16 LUFS for podcast apps.
SECTIONS
| Section | Length | Use |
|---|---|---|
| Intro pulse | 0:20 | Before she starts speaking |
| Narration bed A | 4:00, loops | Main storytelling |
| Narration bed B | 4:00, loops | When the case turns |
| Tension sting | 0:04 | Before a key reveal |
| Break bed | 0:30 | Ad breaks |
| Outro | 0:45 | Credits |
- Section
- Intro pulse
- Length
- 0:20
- Use
- Before she starts speaking
- Section
- Narration bed A
- Length
- 4:00, loops
- Use
- Main storytelling
- Section
- Narration bed B
- Length
- 4:00, loops
- Use
- When the case turns
- Section
- Tension sting
- Length
- 0:04
- Use
- Before a key reveal
- Section
- Break bed
- Length
- 0:30
- Use
- Ad breaks
- Section
- Outro
- Length
- 0:45
- Use
- Credits
LOOPS
Beds A and B loop cleanly. No click at the loop point; I checked each one 10 times in a row.
FILES
- tc_intro_0m20s_-16LUFS.wav
- tc_bedA_loop_4m00s_-34LUFS.wav
- tc_bedB_loop_4m00s_-34LUFS.wav
- tc_sting_0m04s_-20LUFS.wav
- tc_break_0m30s_-24LUFS.wav
- tc_outro_0m45s_-18LUFS.wav
The bed files are already 18 dB quieter, so you can drop them in at 0 dB and her voice stays on top.
TESTED
Played under a 3-minute clip of her narration on earbuds, a car speaker and a laptop. Her words were clear on all three.
— Rafael
