- Home
- Skills
- Research & Analysis
- AudioLab: audio measurement for your agent
Works with the AI Agents you already use
AudioLab: audio measurement for your agent
AudioLab.tools is an independent measurement lab for speech and production audio.
Free
AudioLab: audio measurement for your agent
Example session with this skill installed
Check these two recordings before I deliver them to the client: rec-01.wav and rec-09.wav. Tell me which one has a problem and why, and how far I can trust each number.
- Read your context and instructions
- Compiled the audiolab
rec-09.wav has a problem: it is hard-clipped. rec-01.wav is clean.
| rec-01.wav | rec-09.wav | |
|---|---|---|
| Length | 7.7 s | 3.7 s |
| Clipped samples | 0 (None) | 2,932 (High) |
| Peak | -3.0 dB | 0.0 dB |
What to do: re-export rec-09 from the source with less gain; clipped samples cannot be repaired afterwards. Clipping here counts runs of at least three samples at the peak, so isolated clipped peaks read zero.
How far to trust these numbers. The same answer carries a status for every measure it returns (43 here: 16 validated, 26 proxy, 1 withdrawn). The ones used above:
| Measure | Status | Basis |
|---|---|---|
| clipping | validated | analytic |
| peak level | validated | analytic |
| duration | validated | analytic |
| sibilance risk | proxy | an approximation: good for comparing two recordings, not for stating a fact |
| room echo score | withdrawn | its claim failed its test: it did not measure reverberation, so do not read it as that |
So the clipping verdict is a number you can stand behind; the proxy one is for comparing, and the withdrawn one is not to be used.
Example run through the AudioLab voice-quality tool (analyze_voice, MCP server 0.5.4, 16 November 2026). Both files are public Dutch recordings (FLEURS-NL, CC BY 4.0); we added the clipping to rec-09 ourselves so the result is visible.
Connects securely to your tools. The creator never sees your data.
About this skill
AudioLab.tools is an independent measurement lab for speech and production audio. This skill gives your agent its lenses: 159 ways to measure a recording, each with a published status, so the agent can say how far to trust a number.
What your agent can measure
Free (keyless trial or free key):
- Loudness and levels: integrated loudness, true peak, loudness range, crest factor, and checks against delivery targets (podcast, broadcast, streaming platforms).
- Voice recordings: speech share, signal-to-noise, noise floor, clipping and the background behind the voice.
- Stereo, spectrum and tonal balance, and A/B between two versions.
- Content type, and a check whether a file was upsampled.
- Batches of up to 20 files and files up to 150 MB (free key).
Paid plans add ten named profiles that each answer one question: voice, master, provenance (where the bandwidth really ends, telephone origin, lossy sources), dataset QA, environment, broadcast, loops and packs, room (reverberation from a clap or sweep), chain (dropouts, echo, slapback) and pitch. Pricing: https://audiolab.tools/pricing?utm_source=agensi&utm_medium=listing&utm_campaign=skill
How far to trust it
AudioLab has 159 lenses. 97 are validated against tests; the others are labelled Proxy or Beta, so you always see how far to trust a result. Every answer from the MCP server carries the status of each measure (validated, proxy, beta or withdrawn), with its basis and method. The measurements are rule-based and every result traces back to a stated criterion, so you can explain and justify it. Our test results, misses included, are public: https://audiolab.tools/evidence/studies?utm_source=agensi&utm_medium=listing&utm_campaign=skill
How it connects
The skill uses the AudioLab MCP server (@audiolabtools/mcp-server on npm), which calls the AudioLab API at audiolab.tools. It works with any MCP client, for example Claude Desktop and Claude Code, Cursor, VS Code and Windsurf. It works without a key for light use (20 calls a month, files up to 4 MB). A free key gives 1,000 calls and 2 GB a month: https://audiolab.tools/connect?utm_source=agensi&utm_medium=listing&utm_campaign=skill
Every option, how to read it and what it is not (A, B, C for each): https://audiolab.tools/docs/options?utm_source=agensi&utm_medium=listing&utm_campaign=skill
What leaves your machine
The first time the MCP server starts, npx downloads @audiolabtools/mcp-server from the npm registry (nothing about your audio is sent then). Each call posts the URL, or the file bytes plus the file name (no folder), to audiolab.tools, with your API key if you use one. Files over 4 MB (key needed) go through a signed upload to temporary storage that is cleaned up within the hour. There is no telemetry.
Honest limits
- The browser check misses clipping only for AAC, M4A, OGG and WebM files at 16 kHz; WAV, FLAC and MP3 are read at their own rate.
- The room profile measures reverberation from a clap or sweep, not from speech.
- Telephony origin is a likelihood derived from the bandwidth and does not name a codec.
- One lens (content type) uses a small learned model; a newer version of it (v2) is opt-in in the API until it becomes the default on 15 January 2027.
- It measures. It does not judge naturalness, pronunciation or the words, and it does not change your audio.
The measurement engine runs about 12 times faster than real time on one CPU core of a desktop PC (our own measurement, 21 October 2026); a hosted call adds the upload.
No transcription or tempo detection. Files up to 150 MB; loudness routes up to about 3 hours, voice, signal and profile routes about 7 minutes. The browser check misses clipping in 16 kHz AAC, M4A, OGG and WebM files. Reverberation is measured from a clap or sweep, not from speech. Telephony origin is a likelihood from the bandwidth and names no codec.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed today
- Free with an Agensi account
Requires
Permissions
Allowed Hosts
Network access to audiolab.tools only. It reads only the files you pass to it and sends them (or the URL) to audiolab.tools for analysis. The optional API key is read from the AUDIOLAB_API_KEY environment variable. It does not write files.