Advanced Settings

Fine-tune Intervu's behavior with advanced configuration options.

Overview

Advanced settings affect performance, accuracy, and generation behavior. Most users can leave these at defaults.

---

Audio Processing

Silence RMS Threshold

Audio below this level is treated as silence and not sent to STT.

• Default: 0.005
• Range: 0.001 - 0.1
• Effect: Lower = more audio sent, Higher = more filtering

When to adjust:

• Increase if you hear static/noise being transcribed
• Decrease if quiet speech is being ignored

Audio Chunk Duration

Length of audio chunks sent to STT endpoint.

• Default: 3 seconds
• Range: 1 - 10 seconds
• Effect:
  • Lower = faster initial response, more API calls
  • Higher = fewer API calls, slower response

Recommendation: Keep at default for real-time feel.

---

Transcript Processing

Hallucination Phrases

Whisper sometimes "hears" phrases in silence. These are filtered out.

• Default: you,thank you,thanks,thanks for watching,thank you for watching,thanks for listening,thank you for listening,bye,goodbye,the end,so,okay,hmm,uh,um,oh,ah,i
• Format: Comma-separated list

Adding custom phrases:

you,thank you,thanks,[your custom phrases here]

When to modify:

• If you see recurring false transcriptions
• If certain phrases appear when no one is speaking

Click the Edit in new window button next to the label to open a larger editor for managing your phrase list.

---

LLM Generation

Max Tokens

Maximum tokens in generated answers.

• Default: 0 (unlimited)
• Range: 0 - 8192 (model dependent)
• Effect:
  • Lower = shorter, more concise answers
  • Higher = longer, more detailed answers

Recommendation: Keep at default. Use system prompt to control length.

LLM Debounce Seconds

Delay after interviewer speaks before generating an answer.

• Default: 2 seconds
• Range: 0 - 10 seconds
• Effect:
  • Lower = faster response, may interrupt ongoing speech
  • Higher = slower, but more complete questions

Recommendation: Increase if answers are generated before questions are complete.

LLM Thinking Mode

For models that support extended reasoning (Claude, some local models).

• Default: Off
• Effect: Model may emit thinking tokens before the answer
• Use case: Complex questions requiring reasoning

Queue Mode

Controls how multiple questions are handled.

Mode	Behavior
Off (Priority)	Cancel previous answer, generate for latest question
On (Queue)	Queue all questions, answer in order received

• Default: Off (Priority mode)
• Use Queue mode when: You want answers to all questions in order

Show Queued Answers

When Queue mode is on, show cards for queued questions.

• Default: Off
• Effect: Shows placeholder cardsQuestions waiting for answers
• Use case: Visual feedback that questions are being processed

---

Speaker Diarization

Multi-speaker identification for panel interviews.

Enabling Diarization

Toggle Speaker Diarization in Advanced Settings. The system validates your STT backend supports diarization before enabling.

Configuration

Setting	Description	Default
Speaker Diarization	Enable multi-speaker detection	Off
Min Speakers	Minimum speakers to detect (2-6)	2
Remember Speakers	Save voice profiles across sessions	Off

Min Speakers

Sets the minimum number of speakers the STT backend should look for:

• 2 — Standard interviews (default)
• 3-4 — Small panels
• 5-6 — Large panels or group interviews

Remember Speakers

When enabled, speaker voice embeddings are saved to speaker-profiles.json and restored on the next session. This enables cross-session speaker recognition — the same speaker will be identified consistently across multiple interviews.

Audio Chunk Duration

When diarization is enabled, audio chunk duration is automatically set to a minimum of 6 seconds for better speaker separation. The setting is locked in the UI with an explanatory note.

AI Speaker Naming

When diarization is enabled, the LLM can automatically detect speaker names from conversational cues (e.g., "Hi Anna", "John, what do you think?"). Suggestions appear as sparkle hints next to speaker labels.

See Speaker Diarization Usage for complete details.

---

Interview Timer

Track interview duration and speaking ratios.

Configuration

Setting	Description	Default
Interview Timer	Enable timer display	On
Time Warnings	Show notifications at intervals	Off
Warning Minutes	Comma-separated minutes (e.g., "30,45,60")	30,45,60

Time Warnings

Set intervals at which Intervu will show notifications:

30,45,60  — Notify at 30, 45, and 60 minutes
20,40     — Notify at 20 and 40 minutes
15,30,60  — Notify at 15, 30, and 60 minutes

Validation: Must be comma-separated positive integers in ascending order, each under 1440 (24 hours).

Speaking Ratio

When the timer is enabled, hover over the timer display to see:

• You — Your speaking time and percentage
• Them — Interviewer(s) speaking time and percentage

This helps you balance participation in the interview.

See Interview Timer Usage for complete details.

---

Advanced Mode

Dual-LLM pipeline for intelligent question extraction.

What It Does

Normal mode generates answers whenever the interviewer speaks. Advanced mode:

1. Captures interviewer speech
2. Uses a second LLM to extract complete questions
3. Only generates answers for actual questions

Configuration

Setting	Description
Advanced Mode	Enable/disable
Extractor System Prompt	LLM for question extraction (can use same as main LLM)
Extractor Model	Model for extraction
Extractor API Key	API key if required

Click the Edit in new window button next to the Extractor System Prompt label to open a larger editor.

When to Use

• Enable when: Interviewer talks a lot between questions
• Disable when: You want answers for every statement

---

Contextual Answer Tips

Enable structured answer output with key points and confidence.

When Enabled

LLM outputs:

KEY POINTS:
- First key point
- Second key point
- Third key point

CONFIDENCE: 85

ANSWER:
[Full answer text]

This is parsed and displayed as:

• Bullet points above the answer

• Confidence meter

• Default: Off

• Use case: When you want quick takeaways before reading full answer

---

FFmpeg Path

Custom FFmpeg binary location.

• Default (Windows): Bundled FFmpeg (automatically downloaded)
• Default (macOS): System FFmpeg (must be installed via Homebrew)

When to Change

• If bundled FFmpeg doesn't work on Windows
• If you have a custom FFmpeg build
• If Intervu can't find FFmpeg on macOS

Example Paths

Windows: C:\ffmpeg\bin\ffmpeg.exe
macOS (Intel): /usr/local/bin/ffmpeg
macOS (Apple Silicon): /opt/homebrew/bin/ffmpeg
Linux: /usr/bin/ffmpeg

macOS Installation

FFmpeg is not bundled on macOS. Install it manually:

brew install ffmpeg

See macOS Setup for more details.

---

Settings Backup

Export Settings

Your settings are stored at:

Windows:

%APPDATA%/intervu/settings.json

macOS:

~/Library/Application Support/intervu/settings.json

You can back up this file to preserve your configuration.

Clear Ratings

In Advanced settings, click Clear Ratings to remove all rated answers (used for in-context learning).

---

Troubleshooting Advanced Settings

Too Much Noise Being Transcribed

1. Increase Silence RMS Threshold
2. Add noise patterns to Hallucination Phrases

Answers Generated Before Question is Complete

1. Increase LLM Debounce Seconds
2. Enable Advanced Mode for question extraction

Answers Too Long/Short

1. Adjust Max Tokens
2. Modify System Prompt to specify length requirements

Slack Responses

1. Reduce Audio Chunk Duration
2. Use a faster LLM model
3. Check system resources

---

Next Steps

• Basic Mode Usage
• Advanced Mode Usage
• Speaker Diarization
• Interview Timer