Santhali Speech Data Collection-2

We are currently looking for contributors proficient in the Santhali language for an audio data collection project.

Project Details:

Language: Santhali
Total Effort: 15 hours
Task Type: Audio Data Collection via Gemini App
Activity: Record conversations with the Gemini application

Requirements:

Must be able to read and write in Santhali
Should have basic communication skills in the language
Must have access to 2 mobile devices (required for recording the conversation)
Ability to follow instructions and complete tasks within the given guidelines


Objective behind the task:

Your task is to evaluate the quality of Gemini Live's responses in Indic expansion languages
(Maithili, Santali, Kashmiri, and others). You will be provided with a set of English queries in the
GEMINI Live i18n Testing spreadsheet. For each query, you will translate it into your assigned
target language, conduct a short multi-turn conversation with the model via Gemini Live, and
then rate the model's performance across several quality dimensions in the spreadsheet.
Provided Data
- Query Set: The English prompts are listed in the GEMINI Live i18n Testing spreadsheet
under the "English Prompt" column.
- Target Language / Locale: The language you are assigned to test (e.g., Maithili,
Santali, Kashmiri). Each locale has its own tab in the spreadsheet.
- Rating Sheet: Record all your ratings directly in the spreadsheet tab corresponding to
your locale.
Task
Step-by-Step Process
Step 1: Translate the Query
- Take the provided English query from the "English Prompt" column in the spreadsheet.
- Translate it into your assigned target language. The translation should be natural
and idiomatic — not a word-for-word literal translation.
- Enter the translated prompt in the "Localised prompt" column in the spreadsheet.
- If the query doesn't make cultural sense in the target language, adapt it while preserving
the intent (e.g., replace a US-centric reference with a locally relevant one).
Step 2: Start a Gemini Live Conversation
- Start a new chat for each query. Do not reuse or continue a previous conversation —
each query must be tested in a fresh Gemini Live session to avoid context carryover
from earlier queries.
- Open Gemini Live on your mobile device.

- Speak the translated query in the target language.
- Wait for the model's response.
Step 3: Extend the Conversation (2-3 Follow-up Turns)
- After the model's initial response, continue the conversation for 2-3 additional turns
in the same target language related to the same topic.
- Each turn should have at least a couple of sentences — do not use single-word
follow-ups.
- Your follow-up turns should be natural extensions of the conversation — for example:
- Ask a clarifying question about the response
- Ask the model to elaborate on a point
- Provide a follow-up request related to the original topic
- Correct the model if it responded in the wrong language and see if it adapts
- Stay in the target language throughout the conversation. Do not switch to English or
Hindi unless the model does so first (and in that case, note it in your rating).
Step 4: Share the Conversation Link
After the conversation ends, share the link to the conversation so reviewers can access the
full transcript and context. Follow these steps:
1. Open the Gemini app on your device.
2. Navigate to your conversation history (tap the recent chats / history icon).
3. Find and open the conversation you just completed.
4. Tap the three-dot menu (⋮) at the top of the conversation.
5. Select "Share & export" → "Share".
6. Choose "Get link" or "Copy link" to generate a shareable link.
7. Paste the conversation share link into the "Sharelink" column in the spreadsheet.
Note: Make sure the link is accessible (not restricted to only you). If the share link option is not
available for your conversation, take screenshots of the full transcript (scrolling to capture all
turns), upload them to a Google Drive folder, and paste the Drive folder link instead.
Step 5: Record and Upload the Voice Conversation
Since the conversation link only captures the transcript, we also need the actual voice
recording to evaluate audio quality (pronunciation, accent, TTS). Follow these steps:
1. Before starting the Gemini Live conversation, set up a recording on a separate
device:
- Option A — Second device: Use another phone or laptop to record the audio
from your primary device's speaker.
- Option B — Google Meet: Start a Google Meet on a laptop (just yourself), turn
on recording, then place your phone near the laptop mic and have the Gemini

Live conversation. The Meet recording will capture both your voice and the
model's audio responses.

2. Complete the Gemini Live conversation (Steps 2-3).
3. Stop the recording after the conversation ends.
4. Upload the recording to Google Drive.
5. Paste the Drive link to the recording in the "V