What FreqBlog Is
FreqBlog is a JSON API for music-track metadata: BPM, musical key and mode, energy, danceability, valence, mood, and a number of lower-level audio descriptors — the fields that used to be available via Spotify's /audio-features endpoint before it was retired to third-party developers in November 2024.
We are not a streaming service, and we don't licence, host, or sell audio. The core of the API is descriptive, per-track features — and, derived from those features, a discovery surface (/recommendations, /similar, /related-artists, /radio, /next-track) plus national chart lookups (/charts/{country}). It is designed to be a drop-in replacement for DJ tools, harmonic-mixing apps, recommendation engines, and academic projects that lost access to the Spotify endpoint.
Who Builds It
FreqBlog is operated by Steven Birring, trading as Stackbase / FreqBlog. We are registered with the UK Information Commissioner's Office (ICO) under registration number CSN0293296.
This is a small, independent operation — not a venture-backed company. That is a deliberate choice. It keeps the API independent of streaming-platform commercial pressures and lets us prioritise transparency about coverage and field accuracy over growth-at-all-costs.
How It Works
FreqBlog's analytical fields are computed from a 30-second preview by open-source tooling: Essentia, the audio-analysis library developed by the Music Technology Group at Universitat Pompeu Fabra (Barcelona), detects BPM, key and time signature; librosa measures loudness and the spectral features; and danceability, acousticness, instrumentalness and energy come from models built on PANNs embeddings (Kong et al., 2020, CC BY 4.0), with AcousticBrainz's classifiers preferred where they cover a track. Track identification is keyed on catalogue track IDs, with cross-references to MusicBrainz IDs and ISRC codes where available. Where pre-computed features exist in AcousticBrainz, we incorporate them directly with attribution.
Analysis runs on 30-second commercial preview clips, not full tracks. That is an architectural limit, not a workaround: streaming platforms do not licence full-track audio to third-party services, so any independent provider in this space is preview-based. We are upfront about it because it sets a real ceiling on certain fields — full-track-trained models are typically slightly ahead on mode and key detection for some genres. BPM and tempo-related fields are largely unaffected by clip length, and we publish measurements rather than promises. On our 35-track corroborated reference set (values at least three of four independent annotators agreed on), 85.7% of lookups come back within ±2 BPM. As a broader second opinion we compare against a 2,000-track sample of the Million Song Dataset, The Echo Nest’s 2011 analysis of mostly well-known, pre-2011 recordings. It is a reference point rather than ground truth, and we use it only for this check — never to supply values in our responses. Our BPM agrees with it within ±1 BPM, counting half/double-time as agreement, on 59.2%. Both figures come from our accuracy harness, which re-runs daily; these are from the run of 13 October 2026.
Why We Built It
In November 2024, Spotify deprecated its /audio-features endpoint for third-party developers — a foundational data source that DJ tools, harmonic-mixing apps, BPM-matching tools, recommendation engines, and dozens of academic and indie projects depended on. There has been no comparable open replacement.
FreqBlog is that replacement: independent of any single streaming platform, built on open tooling (Essentia, MusicBrainz, AcousticBrainz), and transparent about its architectural limits. We exist because the gap is real, the use cases are real, and the open-source audio-analysis toolkit is mature enough to do the job credibly.
What We Don't Do
We don't host or stream audio. We don't sell music. We don't sell or share user data with advertisers or analytics brokers (see our privacy policy). We don't pretend to match the field accuracy of full-track-trained models in every case — the trade-off for being independent and preview-based is real, and we'd rather you make an informed integration decision than an over-promised one.
Get in Touch
Questions, feedback, support, partnership enquiries, or bug reports — /contact or [email protected].