voicequal¶

Real-time audio quality assessment for voice apps.

Answers the question every voice app eventually has to answer: "is this recording clean enough to process?"

voicequal analyzes audio with a handful of acoustic metrics, led by the harmonic-to-noise ratio, and returns a tier, excellent, good, fair, or poor, plus the numbers behind the decision.

$ voicequal listen
[15:26:37]  EXCELLENT   room= 45.5 dBA   HNR=18.4dB
[15:26:53]  CHANGE  GOOD   room= 55.0 dBA
[15:26:59]  CHANGE  FAIR   room= 60.4 dBA
[15:27:09]  CHANGE  POOR   room= 71.4 dBA
[15:27:53]  CHANGE  EXCELLENT   room= 45.5 dBA

Install¶

pip install voicequal              # core library
pip install 'voicequal[mic]'       # + live-mic support

Where to go next¶

Quick start: file analysis, streaming, and the CLI in five minutes.
Distilled quality model: DNSMOS-style P.835 scores, pure numpy.
Advice and MCP server: actionable output for people and agents.
How it works: the HNR-gated tier logic and the metrics behind it.
Benchmark: the honest numbers, including the weak category.
API reference and CLI.
Roadmap: where this is going.

Limits, read this before using in production¶

voicequal is calibrated for voice / recording quality. Whether a recording is clean enough to process, not whether it sounds subjectively pleasing to a human.
Not a certified acoustic dB meter. Background dBA is a calibrated proxy using rough dB conversion, not the IEC 61672 A-weighted filter a real SPL meter uses.
Different mics deliver different signal levels. Run voicequal listen --calibrate once per new machine or mic setup.
Assumes reasonable audio input. No echo cancellation or noise suppression built in. If your OS pre-processes mic audio (macOS Voice Isolation, browser noise suppression), your calibration will account for it, but detection accuracy will vary.

License¶

MIT.