MUREKA
Splits a song into 12 individual audio stems and extracts each instrument as MIDI data, returning both stem tracks and MIDI files ready for DAW use.
Avg Run Time: 40.000s
Model Slug: mureka-stem-song-v2-audio-seperation
Playground
Input
Output
Example Result
Preview and download your result.
API & SDK
EACHLABS_API_KEY environment variable. Copy your real API key from /api-keys and set it locally before running.Create a Prediction
Send a POST request to create a new prediction. This will return a prediction ID that you'll use to check the result. The request should include your model inputs and API key.
Get Prediction Result
Poll the prediction endpoint with the prediction ID until the result is ready. The API uses long-polling, so you'll need to repeatedly check until you receive a success status.
Readme
Overview
Mureka | Stem Song V2 | Audio Separation Overview
The Mureka | Stem Song V2 | Audio Separation model from provider Mureka revolutionizes audio processing by splitting full songs into 12 distinct audio stems while extracting MIDI data for each instrument. This solves the challenge of isolating vocals, drums, bass, and other elements from mixed tracks, enabling precise remixing and editing in digital audio workstations (DAWs). Unlike basic stem splitters, its standout feature is simultaneous MIDI extraction, delivering note-level data for instruments like piano, guitar, and synths in two ZIP archives: one for WAV stems and one for MIDI files. Available via the each::labs platform, it supports MP3 or M4A inputs up to ~10 MB, with processing times of 30-90 seconds. Ideal for music producers seeking high-fidelity separation without manual labor, it outperforms simpler tools by providing DAW-ready outputs.
Technical Specifications
Technical Specifications
- Input Formats: MP3 or M4A via public URL or data:audio/mp3;base64,... encoding, up to ~10 MB
- Output: Two ZIP archives—zip_url with 12 individual WAV stems (vocals, bass, drums, piano, guitar, synths, etc.); midi_zip_url with MIDI files for instrument reconstruction
- Stems Supported: 12 high-quality separations including vocals, drums, bass, and 9 additional instruments
- Processing Time: Typically 30-90 seconds per song
- MIDI Extraction: Note data for each instrument, compatible with DAWs like Ableton or Logic Pro
- Architecture: Advanced AI model optimized for multi-stem audio separation and MIDI transcription
Key Considerations
Key Considerations
Before using Mureka | Stem Song V2 | Audio Separation, ensure your audio file is under 10 MB and in MP3 or M4A format; public URLs work best for accessibility. It excels for full songs but may vary in quality with live recordings or heavily compressed tracks. For budget-conscious users, the V1 version (mureka-stem-song) offers 5-stem splits without MIDI at $0.10 per call, trading depth for speed and cost. Optimal for producers needing editable stems and MIDI; test with clean studio mixes first. Access via each::labs API ensures scalable integration into workflows.
Tips & Tricks
Tips and Tricks
Optimize Mureka | Stem Song V2 | Audio Separation by uploading high-bitrate MP3s (320 kbps) for cleaner stem isolation, as lower quality can bleed elements across tracks. Use public URLs from cloud storage like Dropbox for reliable processing. For complex mixes, preprocess with noise reduction tools to enhance separation accuracy. Workflow tip: Import MIDI outputs directly into your DAW to quantize and edit notes before layering new stems.
Example inputs:
- "https://example.com/song.mp3" – Processes a standard pop track into 12 stems + MIDI.
- "data:audio/mp3;base64,/9j/4AAQSkZJRgABAQAAAQABAAD..." – Base64-encoded snippet for quick tests.
- Full album track URL for batch-like single-song analysis.
Combine with each::labs API for automated pipelines, chaining to synthesis models post-separation.
Capabilities
Capabilities
- Splits songs into 12 audio stems including vocals, drums, bass, piano, guitar, synths, and more
- Extracts MIDI files for each instrument, providing note data ready for DAW editing
- Handles MP3 and M4A inputs via URL or base64, up to 10 MB file size
- Delivers dual ZIP outputs: WAV stems and MIDI archives for comprehensive remixing
- Processes full songs in 30-90 seconds with high-fidelity separation
- Supports Mureka | Stem Song V2 | Audio Separation API integration on each::labs
- Reconstructs complex multi-instrument tracks with minimal artifacting
What Can I Use It For?
Use Cases for Mureka | Stem Song V2 | Audio Separation
Music Producers: Isolate vocals and drums from a pop track using "https://example.com/pop-song.mp3" to remix beats in Ableton, leveraging MIDI piano data for custom melodies.
Content Creators: Separate guitar and synth stems from indie songs for YouTube covers; MIDI exports allow pitch-shifting instruments without re-recording, ideal for viral edits.
Developers: Build apps via Mureka | Stem Song V2 | Audio Separation API on each::labs—input user-uploaded tracks to generate karaoke versions by muting vocals.
Sound Designers: Extract bass and percussion from electronic mixes for game audio; MIDI drums enable procedural generation of variations in Unity projects.
Things to Be Aware Of
Things to Be Aware Of
Mureka | Stem Song V2 | Audio Separation performs best on studio-produced tracks; live recordings or podcasts may show stem bleed due to overlapping frequencies. Common mistake: Exceeding 10 MB limits causes failures—compress files first. Base64 inputs suit quick tests but public URLs reduce latency. Edge cases include ambient tracks with minimal instruments, where MIDI may underperform. Monitor processing times, as complex mixes approach 90 seconds. Always verify ZIP downloads promptly.
Limitations
Limitations
Mureka | Stem Song V2 | Audio Separation caps inputs at ~10 MB and MP3/M4A formats only—no WAV or FLAC support. It struggles with non-musical audio like speech or effects-heavy soundscapes, producing inaccurate stems. MIDI extraction assumes standard instrumentation; unconventional sounds yield sparse note data. No real-time processing; batching requires multiple API calls. Quality dips on low-bitrate sources.
Pricing
Pricing Type: Dynamic
Audio Separation V2 (12-stem split + MIDI export): $0.70 per song (Mureka list)
Related AI Models
You can seamlessly integrate advanced AI capabilities into your applications without the hassle of managing complex infrastructure.
Wizper with Timestamp is a multilingual speech recognition and translation model built on Whisper v3 that transcribes audio with precise word-level timestamps. It delivers fast, accurate, and time-aligned transcripts, making it ideal for subtitles, media indexing, and real-time transcription workflows
Wizper is a multilingual speech recognition and translation model based on Whisper v3 that quickly and accurately converts audio files into text. It is optimized for real-time transcription and translation.
xAI Speech-to-Text v1 transcribes audio into text across 25 languages with speaker diarization, word-level timestamps, and clean formatting.
Qwen3-ASR-Flash-Filetrans transcribes audio files into text with support for 26 languages, emotion detection, and word-level timestamps. It is optimized for long audio files (up to 2GB, 12 hours) using asynchronous batch processing. The model supports formats including aac, amr, flac, m4a, mp3, ogg, opus, wav, webm, wma, wmv, as well as video containers. Additional features include inverse text normalization, multi-channel audio transcription, and context biasing for domain-specific vocabulary.
