BANANA PRO AI — CONTROLLED AUDIO CASES
Processed September 16, 2026

These are original synthetic examples, not customer recordings. The artwork is a generated scene illustration; it does not show a real speaker or prove audio quality.

Narration: original scripts, MiniMax Speech 2.6 HD stock Friendly_Person voice, happy emotion, speed 1, pitch 0, volume 1. No real person's voice was cloned. Music: original synthesized sine chords and rhythmic pulses, mixed at 44.1 kHz mono. No third-party music recording was used.

Separation: WaveSpeed audio-vocal-isolator, audio input, default model settings. The vocal output is MP3. Cleaned MP4s were exported by the same browser code used by this tool, copying the original H.264 picture data and encoding the isolated voice to AAC.

Product explainer: 7.745 seconds of audio. Tutorial: 9.579 seconds. Interview illustration: 8.2 seconds. Published separation-price estimates are about $0.0077, $0.0096 and $0.0082 respectively; the API does not return confirmed billed amounts. These provider estimates are separate from Banana Pro AI credits and exclude narration and image generation.

All three source/export video-track hashes match. The last 0.2 seconds of each controlled music-only tail became quieter by 40.5, 19.8 and 32.0 dB at equal gain. This is not a speech-quality score or a promise for your source.

Chrome AAC export introduced approximately 48 ms audio delay and 80–90 ms trailing container padding in these tests. Check lip sync before publishing. Results can contain residual music, altered speech or lost ambience. Singing can remain with the dialogue. Reverb, clipping and overlapping speakers were not validated by these controlled cases.

Rights: original scripts and synthesized music, generated illustrations and a provider stock voice; provider terms apply. Only process material you have permission to use. Music removal does not grant copyright permission.