alt.freeweights.speech / music-whisper
From: laion <huggingface.co/laion>Newsgroups: alt.freeweights.speechSubject:

music-whisper

Date: 2026-11-14Commit: 23dfa36Signed: 2DC28CC7B26A283B ✓

Automatic speech recognition model with 242M parameters. Safetensors weights, 982 MB in all. Published by laion under Apache-2.0.

242M params982MBmusic-whispersafetensorspermissive
FactsParams242MSize982 MBFormatssafetensorsTaskautomatic speech recognitionLicenseApache-2.0From sourcepermissive
$ fw get laion/music-whisper
# manifest sha256 99e0b727c1a5… · 982 MB · 14 files
$ fw verify ./music-whisper laion/music-whisper
From sourceHosted (signed)laion/music-whisperLast check: 2026-11-14 21:34 UTC

Get help: custom fine-tune or private deploy. We review your setup before we quote.

Fine-tune of whisper-small.

Works with

Overview

Size
982 MB
Weights
967 MB (safetensors)
Parameters
242M
Formats
safetensors
Task
automatic speech recognition
License
Apache-2.0 (permissive)
Files
14, each with a sha256
Publisher
laion
Added
2026-11-14
Section
alt.freeweights.speech
Lineage

Original release from the publisher.

Base model: openai/whisper-small

Where these bytes come from
Publisher at the pinned commit→You

Sources are tried in order. A file counts only when its size and sha256 match the manifest. Sources are checked when you ask for them, not on a timer.

Sources
BadgeHostLast checkResult
From sourcehuggingface.co2026-11-14 21:34 UTCpassed

Model card

From the publisher at commit 23dfa36. Shown as plain text with links; images open at the publisher.

Music Whisper (Whisper Small Fine-tune)#

This is a fine-tuned version of OpenAI Whisper Small specifically trained for Music Captioning. Its mean-pooled encoder state also serves as the 768-d audio fingerprint of LAION-Tunes.

Companion release of the NeurIPS 2026 Evaluations & Datasets Track paper LAION-Tunes: An Open 1.4M-Track Dataset and Perceptual Benchmark for AI-Generated Music (Kaczmarczyk, Ahmed, Friedrich, Erickson, Sharoni, Herremans, Schuhmann). Code: LAION-AI/tunes. Demo: Tunes-Search.

Model Details#

  • Base Model: OpenAI Whisper Small
  • Task: Music Captioning (generating textual descriptions of audio tracks)

Training#

Knowledge distillation from Gemini Flash 2.0. The teacher produced six-field structured captions (Overall Sound, Instrumentation, Tempo, Genre, Mood, Vocals) for 2.91M Suno clips of 3 to 30 s from laion/captioned-ai-music-snippets, cut from nyuuzyou/suno.

SettingValue
Base modelopenai/whisper-small (242M parameters), full fine-tune
ObjectiveTeacher-forced cross-entropy over caption tokens
OptimizerAdamW, lr 1e-5, 500 warmup steps, cosine decay
Hardware8x A100, DDP
Effective batch size256 (16 per GPU x 8 GPUs x 2 gradient-accumulation steps)
Released checkpointTraining step 27,218 (validation loss 1.34), more than one pass over the 2.91M clips
CheckpointsEvery 5,000 steps, evaluated on a held-out 200-clip set
Target lengthCaptions truncated to Whisper's 448-token limit
Inferencemax_new_tokens=440

Reference Samples#

This repository includes an HTML file with embedded reference samples that you can download and view to hear the audio alongside the generated captions.

Example Predictions#

Below are examples of captions generated by this model (Ground Truth/Prediction comparison):

Example 1#

"The listener hears a track that immediately establishes itself with a distorted guitar riff, setting a raw and edgy tone. A driving drum beat soon joins, characterized by a prominent snare drum and a heavy kick drum, solidifying the rhythmic foundation. The tempo is approximately 140 beats per minute, contributing to the track's energetic feel. A male vocalist then begins to sing, employing a slightly strained vocal delivery. The vocal profile is that of an adult male, with a timbre and quality described as slightly raspy and aggressive. There are no non-lyrical vocal sounds present. The genre of the music is identified as either Alternative Rock or Hard Rock. The overall mood of the piece is energetic, rebellious, and carries a slight sense of angst. The combination of distorted guitars, the driving drum beat, and the strained vocal delivery are all characteristic elements of these genres. The production quality is relatively clean, yet it retains a raw edge, which enhances the overall intensity of the track. This track would be well-suited for a high-energy action scene, a rock concert, or a rebellious youth anthem."

Example 2#

"The listener hears a piece of music primarily characterized by a slow tempo and a melancholic melody. The primary instrumentation consists of a piano, which carries the main melodic line, and a string section that provides a subtle harmonic foundation. The piano's timbre is described as warm and slightly muffled, contributing to the overall sonic texture. The string section is played with a gentle touch, adding a layer of emotional depth to the piece. The tempo is slow, contributing to the overall mood. The genre of the music is identified as either Classical or Film Score. The mood is predominantly melancholic, reflective, and peaceful. The combination of the piano's melody, the string section's presence, and the string section's delicate sound creates a sense of intimacy and vulnerability. This musical composition would be well-suited for a dramatic scene in a film, a quiet moment of reflection, or as background music for a relaxing activity."

License#

Apache 2.0.

Citation#

@inproceedings{kaczmarczyk2026laiontunes,
  title     = {{LAION-Tunes}: An Open 1.4{M}-Track Dataset and Perceptual Benchmark for {AI}-Generated Music},
  author    = {Kaczmarczyk, Robert and Ahmed, Tawsif and Friedrich, Felix and Erickson, Aidan C. and Sharoni, Orian and Herremans, Dorien and Schuhmann, Christoph},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track},
  year      = {2026}
}

Files

14 files · 982 MBWeights 967 MB (safetensors). Sizes and sha256 from the signed manifest.
Hide all 14 files with their sha256
FileSizesha256
.gitattributes1.52 KB11ad7efa24975ee4b0c3c3a38ed18737f0658a5f75a0a96787b576a78a023361
README.md5.00 KB12dea05a0bbe41522aa72a58962b2453b89bcc5e3f3706da7545300c74d6164e
added_tokens.json34.6 KB9715fd2243b6f06a5858b5e32950d2853f73dd5bc201aafcf76f5082a2d8acd1
config.json1.22 KBb5464429f7e81f7594183985327c7473c0c8811220aa16fcd19b67188ac47921
generation_config.json2.91 KB9a8e1ab969adb3a92b22f62331a0b42050f597ec62ac8112fcdd5ac7312b8b8f
merges.txt494 KB2df2990a395e35e8dfbc7511e08c12d56018d8d04691e0133e5d63b21e154dc6
model.safetensors967 MBd8a08bebb6afd87522f35edf6f8efb134b76aaa684aca582688c688c2b4d2e63
normalizer.json52.7 KBbf1c507dc8724ca9cf9903640dacfb69dae2f00edee4f21ceba106a7392f26dd
preprocessor_config.json356 B994838f1fa6462c8b9b3c90edada831f11f3dd8b4664634e18f4694d005c9dbf
report_step_27218_loss_1.34.html9.31 MB490c38958fbe297c17d2c77e80de8f198fd7d6a337047563bc888b1b9dcd71e4
special_tokens_map.json2.19 KBe67ae3a0aaa99abcd9f187138e12db1f65c16a14761c50ef10eef2c174a7a691
tokenizer.json3.93 MB2d6c907c83c09d556614a8227d264aff2eac2d3f9fc90d9fce4f8fc56a4b9f52
tokenizer_config.json283 KB21a4fc0483c14b87f4e0bbc177a9a357479bfa7c95aaf21ad53d71e9c5afafb9
vocab.json836 KB8f680bba319e01a653d2e8a5dbc17a9157179e0576e6ce74ce0c06356c6e24f9
VersionsEach time the pinned upstream commit or file list changed in this repo. Newest first.
Upstream commitDateContents
23dfa36792 pinned now manifest updated, same commit2026-11-14 21:34 UTC14 files · 982 MB
23dfa367922026-11-14 21:34 UTC14 files · 982 MB

Use it

$ fw get laion/music-whisper
# manifest sha256 99e0b727c1a5… · 982 MB · 14 files
$ fw verify ./music-whisper laion/music-whisper
fw.pyChecks the signature and every sha256, tries each source in order, resumes.
$ python3 fw.py get https://freeweights.allternit.com/manifests/laion/music-whisper.json --dir ./music-whisper $ python3 fw.py verify ./music-whisper ./laion__music-whisper.json
hfThe Hugging Face CLI with one setting. Same repo id.
$ HF_ENDPOINT=https://freeweights.allternit.com hf download laion/music-whisper --revision 23dfa367927599130a7be9cfc33d1205d6e4b427 --local-dir ./music-whisper
transformersSpeech recognition with the transformers pipeline.
import os os.environ["HF_ENDPOINT"] = "https://freeweights.allternit.com" from transformers import pipeline asr = pipeline("automatic-speech-recognition", model="laion/music-whisper", revision="23dfa367927599130a7be9cfc33d1205d6e4b427") print(asr("audio.wav")["text"])

License

Name
Apache License 2.0
Id
Apache-2.0
Class
Permissive
Redistribution
yes
Listing
Hosted (signed)
Decided
License as declared by the publisher. No person read it. 2026-11-14
Text in
README.md
Text sha256
12dea05a0bbe41522aa72a58962b2453b89bcc5e3f3706da7545300c74d6164e
Terms gate
No extra terms beyond the license.

Read the license text

Licenses are checked against the license table. A custom license passes only after a person reads it.

Provenance

From: laion <huggingface.co/laion>Newsgroups: alt.freeweights.speechSubject: music-whisperDate: 2026-11-14Commit: 23dfa36Signed: 2DC28CC7B26A283B ✓
Upstream
laion/music-whisper
Pinned commit
23dfa367927599130a7be9cfc33d1205d6e4b427
Manifest
music-whisper.json · .minisig
Manifest sha256
99e0b727c1a5fc9653ef3ab7977c8159204573f87e5aad7020586fa4f3c09462
Signature
minisign, Ed25519, key 2DC28CC7B26A283B
Created
2026-11-14 21:34 UTC
Generator
freeweights-manifest/0.1.0
Takedowns
None.

How to verify

$ curl -O https://freeweights.allternit.com/manifests/laion/music-whisper.json $ curl -O https://freeweights.allternit.com/manifests/laion/music-whisper.json.minisig $ minisign -Vm music-whisper.json -P RWQ7KGqyx4zCLXtUFOqvfv8L3g+8fpKJ1FsuxPyOYMF+4KufbUxKq1Oq $ shasum -a 256 <file> # compare with the Files tab

A manifest whose signature fails must be rejected. A file whose size or sha256 does not match must be discarded. Verify a download

Model tree

Parent

openai/whisper-small · this model is a finetune of it

1 model in the catalog name this one as their parent.

Fine-tunes
1
Get it
Download
$ HF_ENDPOINT=https://freeweights.allternit.com hf download laion/music-whisper
Open in Freeweights Studio

Pull, verify and run it in the Allternit app on Desktop, web or phone.

Signed manifest
minisign, key 2DC28CC7B26A283B. music-whisper.json. Provenance. Takedown history: none.
From: MIT <huggingface.co/MIT>Subject: ast-finetuned-audioset-10-10-0.4593
From: onnx-community <huggingface.co/onnx-community>Subject: ast-finetuned-audioset-10-10-0.4593-ONNX
From: Edge0 <huggingface.co/Edge0>Subject: Audio8-TTS-Preview-0.6b
From: Edge0 <huggingface.co/Edge0>Subject: Audio8-TTS-Realistic-Air
From: facebook <huggingface.co/facebook>Subject: audiobox-aesthetics
From: mlx-community <huggingface.co/mlx-community>Subject: chatterbox-4bit