@audiolabtools/ai-sdk

@audiolabtools/ai-sdk

Audio analysis as Vercel AI SDK tools: let your LLM check loudness, voice quality, and signal properties of any audio URL, with one import.

npm install @audiolabtools/ai-sdk ai @ai-sdk/anthropic
# (or @ai-sdk/openai, etc.: bring your own model)

import { generateText } from 'ai';
import { audiolab } from '@audiolabtools/ai-sdk';
import { anthropic } from '@ai-sdk/anthropic';

const result = await generateText({
  model: anthropic('claude-opus-4-8'),
  tools: audiolab({ apiKey: process.env.AUDIOLAB_API_KEY }).tools,
  prompt: 'What is the loudness and true-peak of https://example.com/song.wav?',
});

console.log(result.text);

The LLM picks the right tool from your prompt, calls AudioLab's hosted API, and folds the structured response into its answer.

Tools

Every tool takes a public http(s) URL to an audio file. You send only the URL; the AudioLab API downloads the audio and decodes it server-side, so the audio does reach AudioLab (EU-hosted, numbers only come back; see https://audiolab.tools/privacy).

Tool	Endpoint	Returns
analyzeLoudness	POST /mixlab/analyze	Integrated LUFS (EBU R128 / BS.1770-4), true-peak, LRA, crest factor, stereo correlation, tonal-balance bands
checkTarget	POST /mixlab/check-target	Pass/fail vs a delivery target (spotify / apple-music / youtube / tidal / amazon-music / podcast / ebu-broadcast / atsc-broadcast, or custom), with deltas + an ffmpeg loudnorm fix
analyzeTimeseries	POST /mixlab/timeseries	Short-term LUFS over time + downsampled waveform peaks
getSpectrum	POST /mixlab/spectrum	FFT magnitudes + 7-band energies
analyzeVoice	POST /voicelab/qa	Speech/silence ratio, speaking rate, SNR, noise floor, energy in the pauses after speech, sibilance & clipping risk
getSpeechSegments	POST /voicelab/segments	Voiced regions with start/end + per-segment RMS
indexSignal	POST /signallab/index	Content-type guess, tags, clipping/silence regions, brightness & dynamics buckets
compareLoudness	two mixlab/analyze calls	A/B: runs loudness on urlA + urlB, returns both

Full TypeScript types ship in index.d.ts, with autocomplete on every response field.

Configuration

audiolab({
  apiKey?: string;   // AudioLab API key. Falls back to AUDIOLAB_API_KEY env var.
  baseUrl?: string;  // Override base URL. Default: https://audiolab.tools/v1
  fetch?: typeof fetch;  // Custom fetch (for testing / proxying).
})

Use one tool instead of all eight:

const { analyzeLoudness } = audiolab({ apiKey }).tools;
await generateText({ model, tools: { analyzeLoudness }, prompt: '...' });

Status

The hosted API is live with self-serve keys (free tier + paid tiers, Stripe billing live). Sign in at https://audiolab.tools/account to get a key. Prefer to try before wiring it up? The on-device playground at https://audiolab.tools/api#playground runs the same analysis in your browser (no signup, no upload).

License

MIT © Nathan Renting