A lightweight in-browser tool for evaluating microphone signal quality using the Web Audio API. It is designed to help users prepare optimal audio input for speech-to-text models such as Whisper, VOSK, or DeepSpeech — without requiring control over gain settings.
npm i microphone-quality-evaloution@latest
This tool analyzes real-time microphone input and gives simple advice based on:
- RMS (Root Mean Square) energy of your voice
- SNR (Signal-to-Noise Ratio) in the speech frequency band
- Power of the voice signal and background noise
- User behavior (e.g., speaking too softly or loudly)
The output is a short recommendation message, suitable for everyday users (no technical terms), along with a 1–5 mic quality score.
The tool uses a chain of Web Audio nodes for analysis:
-
MediaStreamSource— receives live audio input from the microphone. -
BiquadFilterNode(bandpass) — isolates the frequency range typical of human speech:- Frequency range: 250 Hz to 4000 Hz
- Center frequency:
$\frac{(250 + 4000)}{2} = 2125 Hz$ - Quality factor:
$Q = \frac{center frequency}{bandwidth}$
-
AnalyserNode— uses FFT (size 16384) to measure the frequency-domain energy.
Every 2 seconds:
-
Computes RMS over the target frequency band
-
Estimates power of the voice and background noise
-
Calculates SNR (in dB) using:
$\text{SNR}_{dB} = 10 \cdot \log_{10}\left(\frac{P_{\text{signal}}}{P_{\text{noise}}}\right)$
| Metric | Good Range | Notes |
|---|---|---|
| RMS | ~20 to 70 | Too low = weak voice; too high = possible clipping |
| SNR | > 15 dB (ideal 18+) | Lower SNR indicates high noise or weak speech |
Users are advised to:
- Speak clearly, at a moderate volume
- Move closer to the mic if needed
- Reduce ambient noise (e.g., fans, traffic)
No gain or hardware control is assumed.
The tool displays:
- A brief recommendation message (e.g. “Your voice is too quiet”)
- SNR value and RMS level
- A mic quality score between 1 (poor) and 5 (excellent)
- Pre-checking mic setup for speech recognition
- Ensuring recording conditions are within model-friendly bounds
- Real-time use in web-based voice tools
- Adaptive multi-band filters for more precise isolation
- ML-based noise profiling and environment classification
- User-specific calibration (tracking speaking habits over time)
- Web platform only
- No external JS libraries required
- Fully client-side, secure and lightweight
🧪 Built for experimentation, prototyping, and real-world speech applications.
With ❤️&☕ by codeagha