Tamil-English Bilingual Audio Recording
Budget: $2 – $8 USD
must know tamil and english well for recording audio for 10 usd per hour
RECORDING GUIDELINES
1. What Is This Project?
You are participating in a language data collection project. We are recording natural conversations between two people who speak the same two languages. These recordings will be used to train voice and language technology to better understand how people naturally switch between languages in everyday conversation.
2. Recording Rules
Item
Detail
Sepakers required
You must be 18 years of age or older
You must be a fluent speaker of both languages assigned to your pair
Only native of the required language are allowed to record. Non native recordings will be rejected
Duration per audio
10-15 minutes per audio file
File format:
.wav file, 8 khz or 16kHz, dual channel
Number of Speakers
Conversation must be between 2 roles/2 people only. Do not add more roles in the conversation
Language Use
Use both languages naturally — switch as you normally would in real life
Domains/Topics
Banking, Healthcare, or Travel
Format
Natural conversation — must talk as you normally would.
Natural turn-taking: Speakers should alternate naturally
No silent pause longer than 5 seconds, the flow must be continuously
Try to avoid overlapping conversations
Emotional expressiveness: Emotions should match the context (concern, empathy, frustration etc).
Conversational flow: Dialogue should develop organically, not sound robotic or rehearsed.
3. Recording application set-up:
Record on Zencastr (2 People – 2 Devices at 2 different rooms)
Detailed guidelines: Zencastr_Tool_SOP_.pdf
Download separate tracks for each speaker from the host account.
After recording
Rename files: UUID_subtopic_Language_SpeakerRole
Example: 01_AusEng_Speaker1.wav, 01_AusEng_Speaker2.wav
4. Prepare Before Recording
4.1 Choose a Quiet Environment
Record in a quiet indoor room. Background noise is not allowed
Turn off fans, air conditioners, TVs, radios, phones, and notifications.
No background sounds such as traffic, people talking, pets, music, cracks, clicks from the mouse, or paper rustling, etc.
Your recording must be very clean (SNR > 25, which means speech-to-noise ratio greater than 25)
You must speak loudly enough, avoid heavy breathe into the mic. The mouth should not be close to the mic.
If the audio contains any loud background noise, it will be rejected.
4.2 Check the environment decibel - very important
Download the decibel X application:
For iOS: https://apps.apple.com/us/app/decibel-x-db-sound-level-meter/id448155923
For Android: https://play.google.com/store/apps/details?id=com.skypaw.decibel&hl=en
Step 1: Check Background Noise
Open the Decibel X app.
Stay silent for 10–15 seconds.
Observe the background noise level.
The environment should be around 45 dB.
Step 2: Check Your Speaking Volume
While staying in the same position, speak naturally as you would during recording.
Observe the decibel level while speaking.
Step 3: Ensure Sufficient Volume Difference (SNR)
To meet the requirement SNR > 25 dB, please follow below:
If background noise is 40 dB, your speaking voice should reach ~ 70-75 dB.
If background noise is 45 dB, your speaking voice should reach ~ 75-80 dB.
This difference between background noise and speaking level helps ensure clean audio quality.
Maintain this speaking volume consistently during recording.
Do not shout or force your voice; speak clearly and naturally.
If you cannot reach the required voice level without shouting, please improve the recording environment (quieter room, closer microphone).
Note: Use a good-quality phone/ laptop to record.
Do not use speakerphone or noisy headset mics.
Place the microphone:
About 15–30 cm from your mouth.
Slightly to the side (not directly in front of your mouth).
Avoid touching, bumping, or rustling the microphone or recording device
Keep the microphone at a consistent distance - do not move it during the session
Keep the same position throughout the recording later on.
5. Common Reasons for Rejection (Please Avoid)
Background noise, echo, or SNR smaller than 25dB
NON-NATIVE ACCENT of the target language
Voice clearly sounds like SCRIPT-READING, unnatural tone, or robotic delivery
Wrong file format or sample rate
Audio processing applied
Non-human voice
Wrong names and genders during the conversation (due to reading the reference script and did not change as instructed)
Pause longer than 5 seconds
6. Generate the script (for sample, you dont have to do this)
You can use Claude / GPTchat for help. You can simply create a free account and use it. Recommend to use Claude 4.0 - quite good.
Use the prompt HERE
RECORDING GUIDELINES
1. What Is This Project?
You are participating in a language data collection project. We are recording natural conversations between two people who speak the same two languages. These recordings will be used to train voice and language technology to better understand how people naturally switch between languages in everyday conversation.
2. Recording Rules
Item
Detail
Sepakers required
You must be 18 years of age or older
You must be a fluent speaker of both languages assigned to your pair
Only native of the required language are allowed to record. Non native recordings will be rejected
Duration per audio
10-15 minutes per audio file
File format:
.wav file, 8 khz or 16kHz, dual channel
Number of Speakers
Conversation must be between 2 roles/2 people only. Do not add more roles in the conversation
Language Use
Use both languages naturally — switch as you normally would in real life
Domains/Topics
Banking, Healthcare, or Travel
Format
Natural conversation — must talk as you normally would.
Natural turn-taking: Speakers should alternate naturally
No silent pause longer than 5 seconds, the flow must be continuously
Try to avoid overlapping conversations
Emotional expressiveness: Emotions should match the context (concern, empathy, frustration etc).
Conversational flow: Dialogue should develop organically, not sound robotic or rehearsed.
3. Recording application set-up:
Record on Zencastr (2 People – 2 Devices at 2 different rooms)
Detailed guidelines: Zencastr_Tool_SOP_.pdf
Download separate tracks for each speaker from the host account.
After recording
Rename files: UUID_subtopic_Language_SpeakerRole
Example: 01_AusEng_Speaker1.wav, 01_AusEng_Speaker2.wav
4. Prepare Before Recording
4.1 Choose a Quiet Environment
Record in a quiet indoor room. Background noise is not allowed
Turn off fans, air conditioners, TVs, radios, phones, and notifications.
No background sounds such as traffic, people talking, pets, music, cracks, clicks from the mouse, or paper rustling, etc.
Your recording must be very clean (SNR > 25, which means speech-to-noise ratio greater than 25)
You must speak loudly enough, avoid heavy breathe into the mic. The mouth should not be close to the mic.
If the audio contains any loud background noise, it will be rejected.
4.2 Check the environment decibel - very important
Download the decibel X application:
For iOS: https://apps.apple.com/us/app/decibel-x-db-sound-level-meter/id448155923
For Android: https://play.google.com/store/apps/details?id=com.skypaw.decibel&hl=en
Step 1: Check Background Noise
Open the Decibel X app.
Stay silent for 10–15 seconds.
Observe the background noise level.
The environment should be around 45 dB.
Step 2: Check Your Speaking Volume
While staying in the same position, speak naturally as you would during recording.
Observe the decibel level while speaking.
Step 3: Ensure Sufficient Volume Difference (SNR)
To meet the requirement SNR > 25 dB, please follow below:
If background noise is 40 dB, your speaking voice should reach ~ 70-75 dB.
If background noise is 45 dB, your speaking voice should reach ~ 75-80 dB.
This difference between background noise and speaking level helps ensure clean audio quality.
Maintain this speaking volume consistently during recording.
Do not shout or force your voice; speak clearly and naturally.
If you cannot reach the required voice level without shouting, please improve the recording environment (quieter room, closer microphone).
Note: Use a good-quality phone/ laptop to record.
Do not use speakerphone or noisy headset mics.
Place the microphone:
About 15–30 cm from your mouth.
Slightly to the side (not directly in front of your mouth).
Avoid touching, bumping, or rustling the microphone or recording device
Keep the microphone at a consistent distance - do not move it during the session
Keep the same position throughout the recording later on.
5. Common Reasons for Rejection (Please Avoid)
Background noise, echo, or SNR smaller than 25dB
NON-NATIVE ACCENT of the target language
Voice clearly sounds like SCRIPT-READING, unnatural tone, or robotic delivery
Wrong file format or sample rate
Audio processing applied
Non-human voice
Wrong names and genders during the conversation (due to reading the reference script and did not change as instructed)
Pause longer than 5 seconds
6. Generate the script (for sample, you dont have to do this)
You can use Claude / GPTchat for help. You can simply create a free account and use it. Recommend to use Claude 4.0 - quite good.
Use the prompt HERE