This project aims to collect high-quality Marathi speech recordings to support AI model training and language benchmarking. Participants will record scripted phrases in Marathi under strict quality and technical guidelines.

Role: Marathi Speech Data Collection Participant / ContributorKey Responsibilities
Record 235 Marathi phrases provided by the client using a mobile recording application
235 Marathi phrases includes 118 in task 1 and 117 in task 2
Ensure all the 235 recordings are approved
Follow all recording guidelines strictly (clarity, pronunciation, and pacing)
Record in a completely quiet workplace with no background noise
Maintain minimum 1.5 seconds of silence at the beginning and end of each recording
Ensure accurate reading of provided text (no missing, extra, or repeated words)
Submit recordings as per defined process and timelines
Submit recordings after listening to each one of it. Rerecord in-case of any minor mistake.
Provide correct personal metadata (age, gender, location, device details)

Eligibility Criteria
Native and fluent Marathi reader and speakers only




Age: 18 to 60 years
Balanced gender participation (50% Male / 50% Female ±5%)
Ability to follow instructions and record clearly

Quality Expectations
Transparent pronunciation and natural speech
No background noise or any other human noise or distortion
No speech errors (mumbling, hesitation, or unnatural pauses)
Strict match between text and audio

Rejection Criteria

Recordings or participants may be rejected if:
Non-native speech or incorrect pronunciation
Background noise or poor recording quality
Any other human noise apart from the speaker
Not completed the 235 phrases
Robotic reading of the phrases
Text mismatch (missing/ or adding extra words)
Multiple speakers voice coming from the recording
Failure to meet minimum approved recordings
Unnecessary pause between the phrase
Background noise impacting the speech heavily

Pay: ₹400.00 per hour

Work Location: In person

📌 Audio Moderator Maharashtra
🏢 SpocitServiceBPO
📍 Maharashtra