Models

AudioShake’s Models define the specific type of audio processing applied to your content.
Each model represents a distinct audio processing operation—such as isolating vocals, removing background music, or transcribing dialogue—and can be combined within a single Tasks API request to produce multiple outputs from the same source file.

Models are organized by use case:

• Instrument Stem Separation — break down songs into individual components like vocals, drums, and bass.
• Dialogue, Music, and Effects — isolate voices or remove background elements for film, TV, and dubbing.
• Transcription and Alignment — convert spoken content into synchronized text and timestamps.

Use these models to design flexible workflows for music production, post-production, accessibility, and AI data preparation.

---

Instrument Stem Separation

These models isolate or extract musical components from a mixed track.
They’re useful for remixing, immersive audio, gaming, and music education.
All models can be called via the /tasks route and support standard formats like WAV, MP3, or FLAC.

Name	Model Key	Description	Credits / Minute	Max Length
Instrumental	instrumental	Generates an instrumental-only version by removing vocals. For best quality, use the high_quality variant.	1.0	3 Hours
Drums	drums	Isolates percussion and rhythmic elements.	1.0	3 Hours
Vocals	vocals	Extracts vocal elements from a mix. Supports the high_quality variant for improved clarity.	1.0	3 Hours
Bass	bass	Separates bass instruments and low-frequency sounds.	1.0	3 Hours
Guitar	guitar	Isolates guitar stems (acoustic, electric, classical).	1.0	3 Hours
Piano	piano	Extracts piano or keyboard instruments.	1.0	3 Hours
Strings	strings	Isolates orchestral string instruments like violin, cello, and viola.	1.0	3 Hours
Wind	wind	Extracts wind instruments such as flute and saxophone.	1.0	3 Hours
Other	other	Captures remaining instrumentation after main stems are removed.	1.0	3 Hours
Other-x-Guitar	other-x-guitar	Residual instrumentation after removing vocals, drums, bass, and guitar.	1.0	3 Hours

Residual Stems

To include a residual stem in your results, set "residual": true in the target metadata when creating your task. For more info, contact support@audioshake.ai

Example — Using Models in a Tasks API Request

curl -sS -X POST "https://api.audioshake.ai/tasks" \
  -H "x-api-key: $AUDIOSHAKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com/audio/session.wav",
    "targets": [
      { "model": "vocals", "formats": ["wav"], "variant": "high_quality" },
      { "model": "instrumental", "formats": ["wav"] },
      { "model": "transcription", "formats": ["json"], "language": "en" }
    ]
  }'

---

Dialogue, Music, & Effects

Name	Model	Description	Credits / Minute	Max Length
Dialogue	dialogue	Isolates speech or vocals from any other sound	1.5	3 Hours
Effects	effects	Removes dialogue and music but retains the ambience, sound effects, and environmental noise	1.5	3 Hours
Music removal	music_removal	Removes music from audio while retaining dialogue, background effects, and natural sound	N/A	1 Hour
Background (Music & FX)	music_fx	Removes dialogue to extracting a clean background stem of music and effects	1.5	3 Hours
Music detection	music_detection	Detects the portions of an audio file that contain music	0.5	3 Hours
Multi-Voice	multi_voice	Separates dialogue from multiple speakers in audio recordings, delivering individual audio files per speaker. Available in two_speaker and n_speaker variants, detailed below.	N/A	1 Hour

Music Removal & Multi-Voice Availability

Currently Music Removal and Multi-Voice separation are not available via the /tasks route. Please contact support@audioshake.ai for access.

---

Transcription & Alignment

Name	Model Name	Description	Credits / Minute	Max Length
Transcription	transcription	Text representation of spoken words or audio content	1	1 Hour
Alignment	alignment	Synchronization of audio and corresponding text or captions	1	1 Hour

Combined T&A pricing

If you run Transcription and Alignment together (T&A), pricing is Premium at 1.5 credits per minute.

Example — Transcription and Alignment in a Single Task

You can run both Transcription and Alignment together within one /tasks request.
This produces synchronized text output with word-level timestamps.

curl -sS -X POST "https://api.audioshake.ai/tasks" \
  -H "x-api-key: $AUDIOSHAKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com/audio/interview.mp3",
    "targets": [
      { "model": "transcription", "formats": ["json"], "language": "en" },
      { "model": "alignment", "formats": ["json"], "language": "en" }
    ]
  }'

---

Variants

Certain models offer variants optimized for specific audio processing use-cases. To use a variant, include "variant": "<desired_variant>" in the metadata parameters when submitting a job via the API. The available variants are listed below:

Model	Variant	Description	Plan	Credits / Minute
multi_voice	two_speaker	Optimized for separating two speakers. (Default)	Premium	N/A
multi_voice	n_speaker	Creates stems for any number of speakers.	Advanced	N/A
vocals	high_quality	Higher quality but longer processing time.	Premium	1.5
instrumental	high_quality	Higher quality but longer processing time.	Premium	1.5

Example — Including a Variant in a Tasks API Request

curl -sS -X POST "https://api.audioshake.ai/tasks" \
  -H "x-api-key: $AUDIOSHAKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://example.com/audio/interview.wav",
    "targets": [
      { "model": "multi_voice", "variant": "n_speaker", "formats": ["wav"] }
    ]
  }'