​
Choose parameters by use case

  • Model: LARK_V2 for a polished studio sound; use FINCH (Finch 2) for natural isolation when ambience matters.
  • Loudness: -16 LUFS for stereo or -19 LUFS for mono (common podcast targets).
  • True peak: -1 dBTP (or -2 dBTP if exporting MP3/AAC to avoid intersample peaks).
  • Enhancement level: 80–90 for strong cleanup with natural room; 100 for fully cleaned voice.
  • Transcode: WAV as master; MP3 for distribution.

​
Platform loudness targets (guidance)

  • Podcasts: −16 LUFS (stereo), −19 LUFS (mono); true peak −1 dBTP recommended.
  • YouTube / streaming: around −14 LUFS; true peak −1 dBTP.
  • Music platforms (Spotify/Apple Music): typically normalize to around −14 LUFS; for speech‑first content, −16 to −14 LUFS works well.
  • Broadcast (EBU R128): −23 LUFS with gating; true peak −1 dBTP. Use this when delivering to broadcast specs.
Loudness policies change over time. Treat these as working targets and verify with each platform’s latest guidance.

​
Peak loudness and headroom

  • Set true peak to −1 dBTP by default.
  • Prefer −2 dBTP when exporting to MP3/AAC to reduce intersample clipping risk.
  • Leave at least 1 dB headroom if further mastering or loudness‑normalization is expected downstream.

​
Enhancement level guide

  • 40–60: Subtle cleanup; preserves environment and room tone.
  • 70–90: Strong cleanup while retaining some ambience. Good general‑purpose range.
  • 100: Fully cleaned, voice‑forward result with minimal environment.
If the voice sounds “over‑processed” or brittle, reduce enhancement level or switch from LARK_V2 to FINCH (Finch 2) for more natural isolation.

​
Preset examples

curl -sS -X POST "https://api.ai-coustics.io/v2/medias" \
  -H "X-API-Key: $AICOUSTICS_API_KEY" \
  -F file=@"./episode.wav" \
  -F enhancement='{ "enhancement_model": "LARK_V2", "enhancement_level": 85, "loudness_target": -16, "true_peak": -1, "transcode": "WAV" }'