A "back of the envelope" evaluation intended to answer two questions I have about ASR/STT performance.

Dataset: danielrosehill/ASR-WPM-And-Background-Noise-Eval

• Results Table - Full tabulated results sorted by Word Error Rate
• AI Analysis - In-depth analysis of WPM correlation, background noise impact, and language contamination
• PDF Report - Complete report with visualizations
• CSV Export - Raw data for your own analysis
• Visualizations - Individual charts and graphs

Create annotated audio recordings to evaluate how different factors affect Whisper transcription accuracy:

• Speaking pace (fast, normal, slow, mumbled, whispered)
• Background noise (cafe, office, market, conversations, traffic, wind)
• Microphone distance
• Different microphones

# Create virtual environment and install dependencies
uv venv .venv
source .venv/bin/activate
uv pip install -r requirements.txt

./run.sh
# or
source .venv/bin/activate && python recorder.py

1. Select sample text and microphone
2. Choose annotations (pace, distance, background noise)
3. Add optional notes (noise source URL, volume level, etc.)
4. Record, review, then Save or Discard

Speaking Pace:

• As fast as possible
• Quicker than normal
• Normal/conversational
• Careful enunciation
• Deliberately slow
• Mumbled/unclear
• Whispered
• As loud as possible
• Weird/altered voices

Mic Distance:

• Close (< 6 inches)
• Normal (6-12 inches)
• Far (> 12 inches)

Background Noise:

• Silence
• Coffee shop/cafe
• Busy office
• Busy market
• Background music
• Conversation (same/other/mixed language)
• Traffic/street
• Wind (outdoor)
• Other (see notes)

Each recording saves:

• {id}.wav - 16kHz mono audio (4-char hex ID)
• {id}.json - metadata with all annotations

Example metadata:

{
  "id": "a3f2",
  "sample": "sample_01_technology",
  "word_count": 138,
  "duration_seconds": 62.5,
  "annotations": {
    "pace": "normal",
    "mic_distance": "close",
    "background_noise": "cafe",
    "notes": "imissmycafe.com at 50% volume"
  },
  "equipment": {
    "microphone": "Samson Q2U Microphone",
    "sample_rate": 16000,
    "channels": 1
  }
}

.
├── recorder.py         # Recording GUI
├── run.sh              # Launcher script
├── requirements.txt    # Python dependencies
├── samples/            # Text samples to read
└── dataset/            # Saved audio + metadata