AIVoiceSeparator vs AudioShake

AIVoiceSeparator separates songs into vocal and instrumental stems for backing tracks or clean vocals. Its Studio pipeline combines BS-Roformer, Mel-Roformer, and MDX23C, then applies EBU-R128 loudness normalization. The maker reports an SDR of 12.97 dB and says this is roughly 3 dB above the open Demucs baseline. The web app accepts audio and video files up to 100 MB and 20 minutes in formats including MP3, WAV, M4A, FLAC, OGG, WebM, Opus, and MP4/MOV. You can submit YouTube or SoundCloud links; TikTok clips must be saved and uploaded. Optional Whisper transcription returns SRT, LRC, and TXT files and adds about 30 seconds to processing. A REST API accepts a URL or upload and returns both stems. The free plan allows three songs per month when signed in; pay-as-you-go credits cost $0.50. The maker says uploads and results are deleted within 24 hours and audio is not used to train AI models.

AudioShake separates and processes audio for creators and developers. Studio can split music into vocal and instrument stems or divide finished mixes into dialogue, music and effects. Multi-Speaker separates overlapping voices into labelled tracks, while Speech Recovery denoises and de-reverbs recordings to improve speech intelligibility. The browser-based Lyric Editor transcribes lyrics, aligns words with recordings and supports editing and realignment. Studio accepts WAV, AIFF, FLAC, MP3, AAC, M4A, MP4 and MOV files up to 2 GB. Developers can use REST API processing for batch or streaming workflows, with webhooks; a Local Inference SDK is offered for iOS, macOS, Android, Windows and Linux. Private inference can run in a customer's cloud or fully offline. The free plan starts with 200 credits, MP3 export and files up to 10 minutes. Starter costs $20 a month and Pro costs $40 a month; Enterprise pricing is custom. AudioShake focuses on lyric transcription, not general speech transcription.