Two tools for live speech-to-document reference matching — ideal for interviewers who want instant context from a candidate's resume and documents as the conversation unfolds.
A polished, fully self-contained desktop app that surfaces relevant sections from a candidate's resume and supplemental documents in real time as they speak — so you always have context without needing to pre-read anything.
| Feature | Detail |
|---|---|
| Live transcription | Runs locally via faster-whisper (tiny → medium models). No API key or internet required. |
| Multi-document support | Load PDF, DOCX, and TXT files simultaneously. Documents are chunked and indexed with TF-IDF for sub-second search. |
| Keyword highlighting | Matched keywords are highlighted in colour inside each reference card so you can skim results instantly. |
| Relevance scores | Each card shows a percentage match so you know how strongly a section relates to what was just said. |
| Debounced search | Results update smoothly 600 ms after speech is detected — not on every word. |
| Manual input | A text box lets you test the search without a microphone. |
| Clean dark UI | Modern two-panel layout (live transcript ∣ document references). |
pip install -r requirements.txt
python interview_assistant.py- Click 📂 Load Documents and select the candidate's resume, cover letter, or any other files (PDF / DOCX / TXT).
- Click ▶ Start Listening.
- As the candidate speaks, relevant document sections appear automatically on the right panel.
faster-whisper— recommended; fast, fully local, no internet.openai-whisper— alternative local option (pip install openai-whisper).SpeechRecognition+ Google — fallback, requires internet.
| Model | Speed | Accuracy | RAM |
|---|---|---|---|
| tiny | fastest | lowest | ~1 GB |
| base | fast | good | ~1 GB |
| small | moderate | better | ~2 GB |
| medium | slow | best | ~5 GB |
The original prototype. Listens via the whisper.cpp ./stream binary and searches a TXT file for relevant noun-matched sections.
To run: set up whisper.cpp according to its README, then:
python Run_auto_search_gui.pyThe test document is set to Alice in Wonderland to avoid copyright issues.