audiolab changelog

version

0.5.1 → 0.5.2

version

0.5.0 → 0.5.1

version

0.4.5 → 0.5.0

https://audiolab.tools/llms.txt

+ - [Level and clipping](https://audiolab.tools/voice-ai-check/level-and-clipping): quiet and clipped recordings through two open recognisers, and what AudioLab reads on the same files; clipping is found, no cost was shown+ - [Codecs and bandwidth](https://audiolab.tools/voice-ai-check/codecs): seven speech codecs through two open recognisers, no clear cost; the band is read, a codec is not noticed+ - [Cut-off at the start and end](https://audiolab.tools/voice-ai-check/trimming): cutting the start of the speech cost a recogniser a lot, the end did not; the level at the edge shows it

version

0.4.4 → 0.4.5

https://audiolab.tools/llms.txt

− - [API](https://audiolab.tools/api): the unified analysis engine as an API. Loudness, voice quality, signal classification, A/B comparison, target verification, time-series for graphs, and FFT spectrum, all behind one key. Ten curated tools on the hosted MCP endpoint (the published npm package, 0.3.3, has nine: all but `analyze_profile`), plus a hosted REST API and a Vercel AI SDK provider. Live client-side playground (no signup, no upload), OpenAPI 3.1 spec, and an EBU Tech 3341/3342 + ITU-R BS− - [Connect](https://audiolab.tools/connect): how to add AudioLab to any AI. A local MCP package (Claude, Cursor, Windsurf, VS Code, the Gemini SDK), a hosted remote MCP endpoint at https://audiolab.tools/mcp (OpenAI Responses API, Gemini hosted, n8n), Custom GPT Actions imported from the OpenAPI spec, or the plain REST API. One Bearer key, the same rights everywhere. ChatGPT Developer Mode connectors are not a route: they accept OAuth or no authentication, not an API key.− - **Honest scope:** the engine is on-device-first and EBU/BS.1770-4 loudness-compliant (see the benchmark on the API page); per-task research-SOTA accuracy is not claimed without a published benchmark.− - **Pro (€9/mo, €79/yr):** everything in Free, all analysis presets, 10,000 API + MCP calls/month (20 GB, 120 req/min), custom targets, and deep analysis (voice QA + signal index) on every upload.− - **Founder (€49 one-time, lifetime):** all current and future Pro features; beta-window only, closes at v1.0.− - **Studio (€29/mo, €249/yr):** everything in Pro, 100,000 API + MCP calls/month (200 GB), 300 req/min, priority support.− - **Enterprise (custom, from €499/mo):** self-hosted licences for data-residency or air-gap requirements.− There is no CLI, no webhook feature, no team seats, no cloud sync and no PDF report. If a− - **DSP correctness:** integrated loudness lands within 0.06 LU of the EBU Tech 3341 expected values (tolerance ±0.1 LU) and loudness range matches Tech 3342 exactly, measured against the signals the standard itself defines, with every case published on audiolab.tools/api. One case beyond the standard's frequencies (12 kHz, samples at −3 dBFS) falls outside tolerance: it recovers 1.93 dB of the 3 dB inter-sample peak and reads −1.07 dBTP where an ideal meter approaches 0. No comparison against f− - **Privacy:** the in-browser labs process audio on-device and never upload it to AudioLab.tools. One exception: HearLab's live captions use the browser's built-in speech recognition, which may send the audio to an online speech service (Chrome does). HearLab's level meter runs on the device. HearLab Log measures on the device too and sends only the numbers, and only to the patient's own practice after explicit consent (see /privacy#hearlab). The hosted API and MCP server necessarily receive the+ - [API](https://audiolab.tools/api): the unified analysis engine as an API. Loudness, voice quality, signal classification, A/B comparison, target verification, time-series for graphs, and FFT spectrum, all behind one key. Ten curated tools, the same set on the hosted MCP endpoint and in the npm package (`@audiolabtools/mcp-server`), plus a hosted REST API and a Vercel AI SDK provider. Live client-side playground (no signup, no upload), OpenAPI 3.1 spec, and an EBU Tech 3341/3342 + ITU-R BS.1770+ - [Connect](https://audiolab.tools/connect): how to add AudioLab to any AI. An npm MCP package that runs on your machine and measures on the hosted API (Claude, Cursor, Windsurf, VS Code, the Gemini SDK; a local file path is uploaded to the API, nothing is measured locally), a hosted remote MCP endpoint at https://audiolab.tools/mcp (OpenAI Responses API, Gemini hosted, n8n), Custom GPT Actions imported from the OpenAPI spec, or the plain REST API. One Bearer key, the same rights everywhere. Cha+ - **Honest scope:** the engine runs on-device in the browser labs and server-side behind the API, and is EBU/BS.1770-4 loudness-compliant (see the benchmark on the API page); per-task research-SOTA accuracy is not claimed without a published benchmark.+ - **Plus (€9/mo, €79/yr):** everything in Free, all analysis presets, 10,000 API + MCP calls/month (20 GB, 120 req/min), custom targets, and deep analysis (voice QA + signal index) on every upload.+ - **Founder (€49 one-time, lifetime):** all current and future Plus features; beta-window only, closes at v1.0.+ - **Studio (€29/mo, €249/yr):** everything in Plus, 100,000 API + MCP calls/month (200 GB), 300 req/min, priority support.+ - **Enterprise (custom, from €249/mo):** SLA and dedicated support. A self-hosted or local setup is not a standard product yet; it is built per case, on request, for data-residency or air-gap requirements.+ There is no CLI (use the HTTP API, the Python package or the MCP server), no webhook feature, no team seats and no PDF report.+ Your history in /account and shareable report links exist; both hold numbers, never audio. If a+ - [Voice AI](https://audiolab.tools/voice-ai-check): the path audio takes to a speech recogniser or voice agent in twelve steps: what goes wrong, what AudioLab measures and how far to trust it, what we tested, what it costs a recogniser, what we do not know; a free check in the browser, a stress test in preparation, and the API+ - [Noise and SNR](https://audiolab.tools/voice-ai-check/noise-and-snr): three studies on how far the signal-to-noise estimate can be off, and what noise costs two open recognisers+ - [A competing voice in the background](https://audiolab.tools/voice-ai-check/background-voice): a small, scoped test of two open Whisper models with noise or a second voice added, and what it cannot tell you+ - [Speech detection](https://audiolab.tools/voice-ai-check/speech-detection): two studies on finding where speech is; the first detector missed every criterion, an improved one is an opt-in+ - [Packet loss](https://audiolab.tools/voice-ai-check/packet-loss): finding gaps of digital silence, a stated lower bound, and what packet loss without concealment cost two recognisers+ - [Room and reverberation](https://audiolab.tools/voice-ai-check/room-and-reverb): reverberation time from a clap or a sweep agrees with real rooms; from speech it cannot be measured+ - [Scene measures and content type](https://audiolab.tools/voice-ai-check/scene-and-content-type): four studies on telling speech from music and noise, and the kind of background; what worked and what failed+ - [Do audio measures predict recogniser errors?](https://audiolab.tools/voice-ai-check/error-prediction): in our own test the SNR measure did not; the scope is stated+ - [Where we were wrong](https://audiolab.tools/voice-ai-check/where-we-were-wrong): the measures that failed their tests, what each test showed and what changed+ - **DSP correctness:** integrated loudness lands within 0.06 LU of the EBU Tech 3341 expected values (tolerance ±0.1 LU) and loudness range matches Tech 3342 exactly, measured against the signals the standard itself defines, with every case published on audiolab.tools/api. One case beyond the standard's own signals (a 12 kHz tone with samples at −3 dBFS, so the true peak only shows between the samples) reads +0.1 dBTP where an ideal meter reads 0, inside the +0.2 / −0.4 dB the standard allows fo+ - **Privacy:** the in-browser labs process audio on-device and never upload it to AudioLab.tools. One exception: HearLab's live captions use the browser's built-in speech recognition, which may send the audio to an online speech service (Chrome does). HearLab's level meter runs on the device. HearLab Log measures on the device too and sends only the numbers, and only to the patient's own practice after explicit consent (see /privacy#hearlab). The hosted API and MCP server necessarily receive the+ - **Security and data processing:** audiolab.tools/security summarises where audio goes, where it runs and how it is protected, as a short page for a supplier review; the data processing addendum (Article 28 GDPR), with the sub-processors, is at audiolab.tools/dpa and is the binding text.+ - **LinkedIn:** [AudioLab.tools](https://www.linkedin.com/company/audiolab-tools/)

description

Loudness (EBU R128 / BS.1770-4), true-peak and voice-quality analysis for local files and URLs. → Loudness, true-peak and voice-quality analysis of URLs and local files via the hosted AudioLab API.

version

0.4.3 → 0.4.4

version

0.4.2 → 0.4.3

version

0.4.1 → 0.4.2