Music & Machine Learning
Yinja Audio is a musician-run annotation studio. We turn audio into precise, structured data for music platforms, AI companies and rights libraries — beat grids, keys, structure, captions and score transcription, delivered to your spec.
LIVE SPECTRUM · YINJA AUDIO
What we do
Tempo, key, beat & downbeat grids, section structure, instrumentation, tags and captions. The structural backbone for search, recommendation and cataloguing.
Human listening panels rate AI music on musicality, coherence, style fit and production quality — with clear, citable scorecards your team can act on.
Detailed note-level transcription for single tracks up to multi-instrument works, delivered as MusicXML, MIDI and PDF.
Every batch is double-checked by a senior ear against your annotation guideline before delivery — fatal errors corrected free of charge, further revisions scoped and agreed before they start.
Creative production
Creative work from the same studio — UI & product sound design, game music & audio, AI-assisted music production, and composition & arrangement. One team, two sides of audio.
How it works
Send us your annotation spec and sample tracks. We align on labels, formats and turnaround.
Send 1–2 tracks — we return a fully annotated sample, on the house. No strings.
Approve the sample, set volumes and pricing on your billing basis — hour, track, item or project.
We build your dedicated roster, run two-ear QC on every batch and deliver on schedule.
Trust & credentials
The Yinja Audio label crew — working producers and DJs who annotate, transcribe and quality-check your audio.
Gold-standard calibration before production starts, dual annotation plus senior review on every batch — delivered with an inter-annotator agreement (IAA) report and revision log.
Individual NDAs signed by every annotator — not just the company. Encrypted transfer, no subcontracting without your written approval, deletion on request.
Scoped and quoted before work starts. 48-hour pilot turnaround, batch cadence agreed up front — fatal-error rework at no extra cost, other revisions scoped before they begin.
Delivered in your format — JSON, CSV, XML, MIDI, MusicXML, PDF. References and sample outputs on request before you commit.
20 years as a working DJ, 11+ years composing. Originals featured on CCTV-3's music programme《为爱而歌》(Song for Love); professional judge on Youku's《超感新电音》(Neo-Electronica); scoring for brands including Ping An Bank, Kweichow Zhenjiu and PandaKind — from club sets to film scores.
One of central China's first Melodic Dubstep DJ / producers. Ranked in DubstepCN's annual Top 50 (2019, 2020); a 2018 single with 8M+ streams on China's NetEase Cloud Music; toured with VAC4201 across China, sharing stages with ANTI-GENERAL and KOMPANY.
References, IAA reports, sample batches and NDA terms — available on request. See sample outputs →
Why Yinja Audio
Annotation isn't a desk job to us — it's the same trained listening we do every day as working DJs and producers.
Style knowledge comes from years of deep listening, not from a mixer. Our DJs have spent decades inside vast catalogues — building the instant recognition of style, mood, timbre and structure that MIR systems depend on.
Formal music training from childhood — solfège and ear training — sharpened by hands-on command of timbre, synthesis, orchestration and mixing. From classical and world music to pop, hip-hop and electronic, our producers were trained across the full style map, not just exposed to it.
Every batch is calibrated against your gold samples, passed through dual annotation and senior review — and delivered with an inter-annotator agreement (IAA) report and revision log, the way your ML team actually audits data.
NDA on request, secure transfer, no subcontracting without your written approval. Your audio stays between you and us.
Start here
Tell us your use case and annotation needs — we reply within 24 hours with next steps and a sample timeline.
contact@yinjaaudio.comSample outputs
Illustrative samples for demonstration. Your deliverables follow your own schema, annotation guideline and formats — calibrated against gold samples before production starts.
{
"tempo_bpm": 128,
"time_signature": "4/4",
"key": "A minor",
"sections": [
{"name": "intro", "start_sec": 0.0, "end_sec": 15.8},
{"name": "drop", "start_sec": 15.8, "end_sec": 47.2}
],
"instrumentation": ["kick", "clap", "synth_pad", "vocal"],
"tags": ["melodic house", "euphoric", "bright"]
}
Tempo, key, beat grids, sections, instrumentation and tags — delivered as JSON, CSV or XML.
Track: sample_track_02.wav Musicality 8.2 / 10 Coherence 7.5 / 10 Style fit 8.0 / 10 Production 7.8 / 10 Note: Hook resolves cleanly into the drop; synth timbre fits the brief; low-end could sit tighter in the bridge.
Every score is backed by a written rationale your team can act on — per track or per hour.
<note>
<pitch>
<step>E</step><octave>5</octave>
</pitch>
<duration>1</duration>
<type>quarter</type>
</note>
Single tracks up to multi-instrument works — delivered as MusicXML, MIDI and engraved PDF.