Overview
These guidelines explain how to record high-quality one-on-one conversational data. The goal is to capture natural, realistic conversations while ensuring:- Clear audio quality
- Consistent structure
- Ethical and privacy-safe recording
- Usable data for AI training and evaluation
General recording principles
- Record natural, unscripted conversations
- Avoid reading from scripts unless explicitly instructed
- Speak as you normally would in real life
- Do not exaggerate pronunciation or pacing
- Focus on clarity, not performance
Recording environment
- Choose a quiet, enclosed space
- Avoid background noise (TVs, music, traffic, fans)
- Close windows and doors
- Silence phone notifications
- Avoid echoing rooms when possible
Microphone and device setup
- Use a headset or external microphone when available (see Recommended Microphones)
- Keep the microphone at a consistent distance
- Avoid touching or adjusting the mic during recording
- Do not record while walking or moving
- Ensure stable internet or local recording
Speaker behavior
- Speak at a natural pace
- Do not rush or slow down unnaturally
- Avoid whispering or shouting
- Allow natural pauses and overlaps
- Do not interrupt excessively
Conversational flow
- Treat the session like a real one-on-one conversation
- Respond naturally to what the other speaker says
- Ask follow-up questions when appropriate
- Avoid long monologues unless contextually natural
Natural speech and disfluencies
Natural speech includes hesitations, false starts, self-corrections, and pauses. These are encouraged — they make the data realistic and valuable for AI training.| Encouraged | Avoid |
|---|---|
| Hesitations (um, uh, er) | Overly polished delivery |
| False starts | Scripted or rehearsed responses |
| Self-corrections | Unnatural perfection |
| Natural pauses | Robotic pacing |
Do not try to sound overly polished. Natural, authentic speech is exactly what we need.
Turn-taking
- Allow the other speaker to finish their thoughts
- Overlap is acceptable if it occurs naturally
- Avoid long silences unless conversationally appropriate
- Do not dominate the conversation
Topics and content
- Stay within the assigned topic or task
- Avoid discussing sensitive personal information
- Do not include private data (addresses, passwords, financial info)
- Avoid hate speech, threats, or illegal content
Emotional tone
- Natural emotional expression is acceptable
- Laughter, excitement, and uncertainty are fine
- Do not exaggerate emotions for effect
- Keep reactions authentic
Accents and dialects
- Speak in your natural accent or dialect
- Do not attempt to neutralize or imitate another accent
- Consistency is more important than standardization
Background speech and noise
- Avoid background conversations
- Pause recording if interruptions occur
- Restart only when the environment is quiet
- Do not intentionally include background speakers
Handling recording errors
If an issue occurs during recording:Privacy and consent
- All participants must consent to being recorded
- Do not mention real full names unless instructed
- Avoid personal identifiers
- Follow all platform and project privacy rules
Things to avoid during recording
The following behaviors contaminate conversational data quality and will likely result in rejection:Do not discuss instructions
Do not discuss instructions
Do not talk about task instructions, guidelines, rules, or best practices during the session. Do not reference how the job works or explain what you are supposed to do.
Start the conversation immediately
Start the conversation immediately
Begin speaking as soon as the prompt or question appears. Do not introduce the task, explain the setup, or announce that you are starting.
Avoid meta commentary
Avoid meta commentary
Do not talk about timers, time limits, recording status, the platform or interface, reading instructions, or “doing a job” or “completing a task.” The conversation should feel natural and self-contained.
Summary checklist
Before and during every recording session, confirm:- Quiet environment with no background noise
- Clear, consistent audio from an approved microphone
- Natural conversational flow — not scripted or rehearsed
- Authentic speech patterns with natural disfluencies
- No sensitive or private information shared
- No forced or scripted delivery
- Proper consent obtained from all participants
- No discussion of instructions, timers, or task mechanics
- Conversation starts immediately from the prompt
For equipment recommendations, see Recommended Microphones and Recording Setup.